0 to over $8M ARR in 2 months as a Claude Wrapper (Bolt.new, Qodo)

00:00:00.000 | (upbeat music)

00:00:02.580 | - Hey, everyone.

00:00:04.940 | Welcome to the Latent Space Podcast.

00:00:06.620 | This is Alessio, partner and CTO at Decibel Partners,

00:00:09.220 | and I'm joined by my co-host, Swiggs, founder of Small AI.

00:00:12.040 | - Hey, and today we're still in our sort of makeshift

00:00:14.840 | in-between studio, but we're very delighted

00:00:17.360 | to have a former returning guest host, Itamar.

00:00:20.640 | Welcome back.

00:00:21.600 | - Great to be here after a year or more.

00:00:23.360 | - Yeah, a year and a half.

00:00:24.640 | You're one of our earliest guests on Agents.

00:00:26.920 | Now you're CEO co-founder of Kodo.

00:00:29.760 | - Right.

00:00:30.600 | - Which has been, it's just been renamed.

00:00:31.600 | You also raised a $30 million Series A.

00:00:34.240 | - Series A was 40.

00:00:35.320 | - 40, $40 million Series A,

00:00:37.000 | and we can get caught up on everything,

00:00:39.160 | but we're also delighted to have our new guest, Eric.

00:00:41.940 | Welcome.

00:00:42.780 | - Thank you. Excited to be here.

00:00:43.800 | - Should I say Bolt or StackBlitz?

00:00:45.280 | Like, is it like its own company now or?

00:00:47.600 | - Yeah, Bolt's definitely Bolt.new.

00:00:50.280 | That's the thing that we're probably the most known for,

00:00:52.520 | I imagine, at this point.

00:00:54.480 | - Which is ridiculous to say

00:00:55.520 | because you were working at StackBlitz for so long.

00:00:57.800 | - Yeah, I mean, it was within a week,

00:00:59.720 | we were doing like double the amount of traffic.

00:01:02.780 | And StackBlitz had been online for seven years,

00:01:04.480 | and we were like, what?

00:01:05.320 | But anyways, yeah, so we're StackBlitz,

00:01:07.600 | the company behind Bolt.new.

00:01:09.440 | If you've heard of Bolt.new, that's our stuff, so.

00:01:11.720 | - Yeah, yeah, excellent.

00:01:13.000 | - I see, by the way, that the founder mode,

00:01:14.400 | you need to know to capture opportunities.

00:01:16.160 | So, like, kudos on doing that, right?

00:01:18.720 | Like, you're working on some technology,

00:01:20.080 | and then suddenly you can exploit that to a new world.

00:01:23.760 | - Totally, yeah, I mean, you know, I think,

00:01:25.840 | well, not to jump, but at least 100%.

00:01:28.600 | I mean, that's, I mean, kind of a couple of months ago,

00:01:31.960 | we had the idea for Bolt earlier this year,

00:01:34.400 | you know, we haven't really shared this too much publicly,

00:01:36.560 | but like, you know, we actually had tried to build it

00:01:38.000 | with some of the state-of-the-art models

00:01:39.720 | back in January, you know, February,

00:01:41.320 | you can kind of imagine which,

00:01:42.640 | and they just like weren't good enough

00:01:44.520 | to actually do the code generation

00:01:46.600 | where the code was accurate and, you know,

00:01:48.840 | it was fast and whatever have you,

00:01:50.000 | without a ton of like rag,

00:01:51.120 | but then there was like issues with that.

00:01:52.500 | So we put it on the shelf,

00:01:53.420 | and then we got kind of a sneak peek of, you know,

00:01:54.760 | some of the new models that have come out,

00:01:56.560 | you know, in the past couple of months now.

00:01:58.640 | And so once we saw that,

00:01:59.760 | once we actually saw the code gen from it,

00:02:01.360 | we were like, oh my God, like, okay,

00:02:03.000 | then we can build a product around this.

00:02:05.040 | And so that was really the impetus of us,

00:02:07.240 | of us building the thing.

00:02:08.360 | But with that, it was, you know, StackBlitz,

00:02:10.200 | the core StackBlitz product the past seven years

00:02:12.080 | has been an IDE for developers.

00:02:14.840 | So the entire user experience flow we've built up

00:02:16.860 | just didn't make sense.

00:02:17.920 | And so when we kind of went out to build Bolt,

00:02:19.480 | we just thought, you know,

00:02:21.040 | if we were inventing our product today,

00:02:23.880 | what would the interface look like,

00:02:25.560 | given what is now possible with the AI code gen?

00:02:27.920 | And so there's definitely a lot of conversations

00:02:29.960 | we had internally, but you know,

00:02:31.480 | just kind of when we logically laid it out,

00:02:33.080 | we were like, yeah, I think it makes sense

00:02:35.640 | to just greenfield a new thing and let's see what happens.

00:02:38.920 | If it works great, then, you know, we'll figure it out.

00:02:41.440 | If it doesn't work great,

00:02:42.280 | then it'll get deleted at some point, right?

00:02:45.140 | But yeah, anyway, so that's kind of the,

00:02:48.720 | how it actually came to be.

00:02:49.920 | - I'll mention your background a little bit.

00:02:51.640 | You were also founder of Thinkster

00:02:53.400 | before you started StackBlitz.

00:02:55.040 | So both of you are second time founders.

00:02:57.160 | Both of you have sort of re-founded your company recently.

00:03:00.560 | Yours was more of a rename.

00:03:02.000 | I think a slightly different direction as well.

00:03:03.520 | And then we can talk about Bolt.

00:03:04.640 | Maybe just chronologically,

00:03:05.760 | should we get caught up on where Codo is first?

00:03:08.880 | And then, you know, just like what people should know

00:03:10.760 | since the last pod?

00:03:12.440 | - Sure, like the last part was two months

00:03:14.160 | after we launched and we basically had the vision

00:03:17.280 | that we talked about.

00:03:18.520 | The idea that software development is about specification,

00:03:21.720 | test and code, et cetera.

00:03:23.800 | We are more on the testing part as in essence,

00:03:26.320 | we think that if you solve testing,

00:03:29.020 | you solve software development.

00:03:30.400 | - The beautiful chart that we'll put up on screen.

00:03:33.040 | - And testing is a really big field.

00:03:35.120 | Like there are as many dimension, unit testing,

00:03:37.920 | the level of the component, how big it is, how large it is.

00:03:41.440 | And then there is like different type of testing.

00:03:43.480 | Is it regression or smoke or whatever?

00:03:45.380 | So back then we only had like a one ID extension

00:03:49.360 | with unit tests as in focus.

00:03:51.920 | So one and a half year later, first ID extension

00:03:55.440 | supports more type of testing.

00:03:57.720 | As a context of where we index local repos,

00:04:01.880 | but also 10,000s of repos for Fortune 500 companies.

00:04:06.280 | We have another agent, another tool that is called,

00:04:10.560 | the peer agent is the open source

00:04:11.920 | and the commercial one is Codo Merge.

00:04:14.520 | And then we have another open source called the cover agent,

00:04:18.200 | which is not yet commercial product coming very soon.

00:04:21.800 | It's very impressive.

00:04:23.120 | It could be that already people are approving

00:04:26.200 | automated pull requests that they don't even aware

00:04:29.520 | in really big open sources.

00:04:31.080 | So once we have enough of these,

00:04:32.840 | we will also launch another agent.

00:04:35.960 | In one and a half year, what we did is grew in our offering

00:04:40.200 | and mostly on the side of does this code actually works?

00:04:44.280 | Testing, code review, et cetera.

00:04:46.160 | And we believe that's the critical milestone

00:04:50.680 | that needs to be achieved to actually have

00:04:52.920 | the AI engineer for enterprise software.

00:04:55.960 | So, and then like for the first year was everything

00:04:59.760 | bottom up getting to 1 million installation.

00:05:02.600 | 2024, that was 2023, 2024 was starting to monetize,

00:05:06.560 | to feel like how it is to make the first buck.

00:05:09.840 | So we did the teams offering,

00:05:11.160 | it went well with a thousand of teams, et cetera.

00:05:13.560 | And then we started like just a few months ago

00:05:16.840 | to do enterprise with everything you need,

00:05:20.280 | which is a lot of things that discussed in the last post

00:05:23.520 | that was just released by Codium.

00:05:26.960 | That's how we call it at Codium.

00:05:30.040 | Just opening the brackets,

00:05:32.320 | our company name was Codium AI and we renamed to Codo

00:05:36.680 | and we call our models Codium.

00:05:38.400 | So back to my point, so we started an enterprise motion

00:05:40.840 | and already have like multiple like fortune 100 companies.

00:05:44.200 | And then with that, we raised a series say of $40 million.

00:05:49.080 | And what's exciting about it is that enables us

00:05:52.840 | to develop more agents.

00:05:55.240 | That's our focus.

00:05:56.080 | Like, I think it's very different.

00:05:57.880 | We're not coming very soon with an ID or something like that.

00:06:01.080 | - You don't want to fork this code?

00:06:02.840 | (laughing)

00:06:03.680 | - Yeah, maybe we'll fork JetBrains or something

00:06:06.480 | just to be different.

00:06:07.400 | (laughing)

00:06:08.480 | - I noticed that, you know, I think the promise

00:06:10.280 | of general purpose agents has kind of died.

00:06:12.360 | Like everyone is doing kind of what you're doing.

00:06:14.760 | There's CodoGen, CodoMerge and then there's a third one.

00:06:16.560 | What's the name of it?

00:06:17.400 | - Yeah, CodoCover.

00:06:18.600 | - CodoCover.

00:06:19.440 | - Which is like a commercial version of a cover agent.

00:06:22.240 | It's coming soon.

00:06:23.080 | - Yeah, I see something similar with factory AI

00:06:25.400 | also doing like droids.

00:06:27.200 | They all have special purpose doing things,

00:06:29.680 | but people don't really want general purpose agents, right?

00:06:32.280 | The last time you were here, we talked about AutoGBT,

00:06:34.160 | the biggest thing of 2023.

00:06:35.960 | This year, not really relevant anymore.

00:06:37.560 | And I think it's mostly just 'cause like,

00:06:39.560 | when you give me a general purpose agent,

00:06:41.160 | I don't know what to do with it.

00:06:42.480 | - Yeah, I totally agree with that.

00:06:44.400 | We're seeing it for a while

00:06:45.440 | and I think it will stay like that

00:06:47.240 | despite the computer use, et cetera,

00:06:49.000 | that supposedly can just replace us

00:06:50.520 | and you can just like prompt it to be,

00:06:52.640 | hey, now be a QA or be a QA person or a developer.

00:06:57.520 | I still think that there's a few reasons

00:06:59.760 | why you see like a dedicated agent.

00:07:02.120 | Again, I'm a bit more focused,

00:07:03.920 | like my head is more on complex software

00:07:07.600 | for big teams and enterprise, et cetera.

00:07:10.080 | And even think about permissions

00:07:13.280 | and what are the data sources

00:07:15.600 | just the same way you manage permissions for users,

00:07:18.760 | like developers, you probably want

00:07:21.000 | to have dedicated guardrails

00:07:22.640 | and dedicated approvals for agents.

00:07:25.120 | I intentionally like touched a point

00:07:26.960 | on not many people think about.

00:07:28.960 | And of course, then what you can think of,

00:07:31.320 | like maybe there's a different tools, tool use, et cetera.

00:07:35.480 | But just the first point by itself

00:07:37.480 | is a good reason why you wanna have different agents.

00:07:40.120 | - Just to compare that with Pod.new,

00:07:42.520 | you're almost focused on like,

00:07:44.600 | the application is very complex

00:07:46.280 | and now you need better tools to kind of manage it

00:07:48.400 | and build on top of it.

00:07:49.560 | On Bolt, it's almost like,

00:07:50.520 | like I was using it the other day.

00:07:52.200 | There's basically like, hey, look,

00:07:53.480 | I'm just trying to get started.

00:07:55.080 | I'm not very opinionated

00:07:56.320 | on like how you're gonna implement this,

00:07:57.960 | like this is what I wanna do.

00:07:59.720 | And you build a beautiful with it.

00:08:01.920 | What people ask as the next step,

00:08:04.160 | going back to like the general versus like specific,

00:08:06.880 | have you had people say, hey, this is great to start,

00:08:09.200 | but then I want a specific Bolt.new.whatever else

00:08:12.760 | to do a more vertical integration

00:08:14.880 | and kind of like development or what do people say?

00:08:18.000 | - Yeah, I think you kind of hit the head on,

00:08:20.480 | which is kind of the way

00:08:21.760 | that we've kind of talked about internally

00:08:23.480 | is it's like people are using Bolt

00:08:25.240 | to go from like 0.0 to 1.0.

00:08:27.760 | Like that's like kind of the biggest unlock that Bolt has

00:08:30.120 | versus most other things out there.

00:08:32.400 | I mean, I think that's kind of,

00:08:33.520 | you know, what's very unique about Bolt.

00:08:35.480 | I think the, you know,

00:08:37.200 | the working on like existing enterprise applications is,

00:08:40.440 | I mean, it's crazy important because, you know,

00:08:42.240 | there's, when you look at the Fortune 500,

00:08:44.440 | I mean, these code bases,

00:08:45.280 | some of these have been around for, you know,

00:08:46.600 | 20, 30 plus years.

00:08:48.480 | And so it's important to be going from, you know,

00:08:49.960 | 101.3 to 101.4, et cetera.

00:08:53.160 | I think for us,

00:08:54.000 | so what's been actually pretty interesting

00:08:55.600 | is we see there's kind of two different users for us

00:08:59.240 | that are coming in and it's very distinct.

00:09:01.000 | It's like people that are developers already.

00:09:03.160 | And then there's people

00:09:04.000 | that have never really written software.

00:09:05.520 | And more if they have, it's been very, very minimal.

00:09:08.000 | And so in the first camp,

00:09:09.160 | what these developers are doing,

00:09:10.160 | like to go from zero to one,

00:09:11.440 | they're coming to Bolt

00:09:12.280 | and then they're ejecting the thing to get up

00:09:14.000 | or just downloading and, you know,

00:09:15.200 | opening cursor, like whatever,

00:09:16.880 | to keep iterating on the thing.

00:09:18.520 | And sometimes they'll bring it back to Bolt

00:09:19.760 | to like add in a huge piece of functionality

00:09:22.080 | or something, right?

00:09:22.960 | But for the people that don't know how to code,

00:09:24.360 | they're actually just, they live in this thing.

00:09:26.560 | And that was one of the weird things when we launched is,

00:09:28.360 | you know, within a day of us being online,

00:09:31.160 | one of the most popular YouTube videos,

00:09:32.800 | and there's been a ton since,

00:09:33.720 | which was, you know, there's like,

00:09:34.600 | oh, Bolt is the cursor killer.

00:09:36.120 | And I originally saw the headlines and I was like,

00:09:38.440 | thanks for the views.

00:09:39.440 | I mean, I don't know, this doesn't make sense to me.

00:09:41.640 | That's not what we kind of thought.

00:09:43.680 | - It's how YouTubers talk to each other.

00:09:46.120 | Everything kills everything else.

00:09:47.560 | - Totally, but what blew my mind

00:09:49.400 | was that there was any comparison.

00:09:50.400 | 'Cause it's like cursor is a local IDE product.

00:09:53.880 | But when we actually kind of dug into it

00:09:55.920 | and we have people that are using our product saying this,

00:09:58.600 | I'm not using cursor.

00:09:59.440 | And we're like, what?

00:10:00.280 | And it turns out there are hundreds of thousands of people

00:10:03.080 | that we have seen that we're using cursor

00:10:06.840 | and we're trying to build apps with that

00:10:08.520 | where they're not traditional software to us,

00:10:09.880 | but we're heavily leaning on the AI.

00:10:12.200 | And as you can imagine,

00:10:13.120 | it is very complicated, right, to do that with cursor.

00:10:16.400 | So when Bolt came out, they're like,

00:10:17.400 | wow, this thing's amazing

00:10:18.920 | because it kind of inverts the complexity

00:10:21.800 | where it's like, you know, it's not an IDE,

00:10:23.840 | it's a chat-based sort of interface that we have.

00:10:27.000 | So yeah, so that's kind of the split,

00:10:28.240 | which is rather interesting.

00:10:29.080 | We've had like the first startups now launch

00:10:30.800 | off of Bolt entirely, where this, you know,

00:10:33.000 | tomorrow I'm doing a live stream with this guy named Paul,

00:10:35.520 | who he's built an entire CRM using this thing

00:10:38.680 | and with backend, et cetera.

00:10:40.320 | And people have made their first money on the internet,

00:10:43.160 | period, you know, launching this with Stripe

00:10:45.160 | or whatever have you.

00:10:46.000 | So anyways, that's kind of the two main categories

00:10:49.120 | of folks that we see using Bolt though.

00:10:51.000 | - I agree that I don't understand the comparison.

00:10:53.760 | It doesn't make sense to me.

00:10:54.600 | I think like we have like two type of families of tools.

00:10:58.320 | One is like we re-imagine the software development.

00:11:00.720 | I think Bolt is there.

00:11:02.560 | And I think like a cursor is more like a evolution

00:11:04.920 | of what we already have.

00:11:06.560 | It's like taking the IDE and it's amazing.

00:11:09.680 | And it's okay, let's adapt the IDE to an era

00:11:13.680 | where LLMs can do a lot for us.

00:11:15.640 | And Bolt is more like, okay,

00:11:17.520 | let's rethink everything totally.

00:11:19.800 | And I think we see a few tools there,

00:11:21.960 | like maybe Vercel, Veo, and maybe Repl.it in that area.

00:11:26.720 | And then in the area of let's expedite, let's change,

00:11:31.440 | let's progress with what we already have.

00:11:33.720 | You can see Cursor and Codo.

00:11:36.040 | We're different between ourself, Cursor and Codo,

00:11:39.080 | but definitely I think like comparison doesn't make sense.

00:11:42.120 | - And just to set the context, this is not a Twitter demo.

00:11:45.440 | You've made 4 million of revenue in four weeks.

00:11:47.640 | So this is actually working, you know, it's not a...

00:11:51.760 | What do you think that is?

00:11:52.800 | Like there's been so many people demoing coding agents

00:11:55.840 | on Twitter and then it doesn't really work.

00:11:57.920 | And then you guys were just like, here you go, it's live,

00:12:00.840 | go use it, pay us for it, you know?

00:12:02.840 | Is there anything in the development

00:12:04.200 | that was like interesting and maybe how that compares

00:12:06.400 | to building your own agents?

00:12:07.960 | - We had no idea, honestly.

00:12:09.520 | Like we've been pretty blown away

00:12:11.520 | and then things have just kind of continued

00:12:12.760 | to grow faster since then.

00:12:14.640 | We're like, oh, today is week six.

00:12:16.440 | So I kind of came back to the point you just made, right?

00:12:18.680 | Where it's, you kind of outlined,

00:12:20.560 | it's like there's kind of this new market

00:12:22.400 | of like kind of rethinking the software development.

00:12:25.240 | And then there's heavily augmenting existing developers.

00:12:28.000 | I think that, you know, both of which are, you know,

00:12:31.200 | AI code gen being extremely good,

00:12:33.880 | it's allowed existing developers,

00:12:35.840 | it's allowing existing developers to hammer out software

00:12:39.080 | far faster than they could have ever before, right?

00:12:41.040 | It's like the ultimate power tool for an existing developer.

00:12:43.640 | But this code gen stuff is now so good.

00:12:45.720 | And we saw this over the past, you know,

00:12:47.600 | from the beginning of the year,

00:12:48.480 | when we tried to first build Bolt,

00:12:50.280 | it's actually lowered the barrier to people

00:12:52.760 | that aren't traditionally software engineers.

00:12:55.120 | But the kind of the key thing is,

00:12:56.440 | if you kind of think about it from,

00:12:57.480 | imagine you've never written software before, right?

00:13:00.520 | My co-founder and I,

00:13:01.400 | he and I grew up down the street from each other in Chicago.

00:13:04.280 | We learned how to code when we were 13 together,

00:13:05.640 | and we've been building stuff ever since.

00:13:07.520 | And this is back in like the mid 2000s or whatever,

00:13:09.840 | you know, there was nothing for free to learn from online

00:13:12.080 | on the internet and how to code.

00:13:13.240 | So like we, for our 13th birthdays,

00:13:14.960 | we asked our parents for, you know, O'Reilly books,

00:13:17.040 | 'cause you couldn't get this at the library, right?

00:13:19.000 | And so instead of like an Xbox, we got, you know,

00:13:20.960 | programming books.

00:13:22.480 | But the hardest part for everyone learning to code

00:13:24.680 | was getting an environment set up locally, you know?

00:13:26.960 | And so when we built StackBlitz,

00:13:28.760 | like kind of the key thesis, like seven years ago,

00:13:31.000 | the insight we had was that,

00:13:33.880 | hey, it seems like the browser has a lot of new APIs,

00:13:36.760 | like WebAssembly and Service Workers, et cetera,

00:13:39.240 | where you could actually write an operating system

00:13:41.200 | that ran inside the browser,

00:13:42.520 | that could boot in milliseconds.

00:13:44.080 | And you, you know, basically there's this missing capability

00:13:47.080 | of the web, like the web should be able

00:13:48.560 | to build apps for the web, right?

00:13:50.720 | You should be able to build the web on the web.

00:13:52.520 | Every other platform has that, Visual Studio for Windows,

00:13:55.560 | Xcode for Mac, the web has no built-in primitive for this.

00:13:58.880 | And so just like our built-in kind of like nerd instinct

00:14:01.520 | on this was like, that seems like a huge hole

00:14:03.680 | and it's, you know, it will be very valuable

00:14:05.400 | or like, you have a very valuable problem to solve.

00:14:07.080 | So if you want to set up dev environments, you know,

00:14:09.040 | this is what we spent the past seven years doing.

00:14:11.120 | And the reality is existing developers running locally,

00:14:14.800 | they already know how to set up dev environments,

00:14:16.120 | so the problem isn't as acute for them.

00:14:17.920 | When we put Bolt online,

00:14:19.040 | we took that technology called WebContainer

00:14:21.560 | and married it with these, you know,

00:14:22.680 | state-of-the-art frontier models.

00:14:24.640 | And the people that have the most pain

00:14:27.760 | with getting stuff set up locally

00:14:28.760 | is people that don't code.

00:14:30.760 | I think that's been, you know,

00:14:32.000 | really the big explosive reason is

00:14:34.480 | no one else has been trying to make dev environments work

00:14:36.760 | inside of a browser tab, you know, for the past,

00:14:39.840 | if since ever, other than basically our company,

00:14:42.200 | largely because there wasn't an immediate demand or need.

00:14:44.680 | So I think we kind of find ourselves

00:14:45.800 | at the right place at the right time.

00:14:47.560 | And again, for this market of people

00:14:48.720 | that don't know how to write software,

00:14:50.800 | you would kind of expect

00:14:51.640 | that you should be able to do this

00:14:53.280 | without downloading something to your computer

00:14:54.960 | in the same way that,

00:14:56.040 | hey, I don't have to download Photoshop now

00:14:57.840 | to make designs 'cause there's Figma.

00:14:59.280 | I don't have to download Word

00:15:00.120 | because there's, you know, Google Docs.

00:15:01.800 | They're kind of looking at this as that sort of thing, right?

00:15:04.600 | Which was kind of the, you know,

00:15:06.040 | our impetus and kind of vision from the get-go.

00:15:08.320 | But, you know, the code gen,

00:15:10.200 | the AI code gen stuff that's come out

00:15:11.560 | has just been, you know,

00:15:13.080 | an order of magnitude multiplier on how magic that is, right?

00:15:16.440 | So it was that kind of my best distillation

00:15:18.240 | of like what is going on here, you know?

00:15:20.360 | - Yeah, and you're going to deploy too, right?

00:15:22.280 | - Yes.

00:15:23.120 | - In TechBlitz, yeah.

00:15:23.960 | - Yeah, and so that's what's really cool is it's, you know,

00:15:25.200 | we have deployment built in with Netlify,

00:15:27.560 | and this is actually, I think Sean,

00:15:30.360 | you actually built this at Netlify when you were there.

00:15:33.440 | One of the most brilliant integrations, actually,

00:15:35.920 | because, you know, effectively the API that Sean built,

00:15:38.280 | maybe you can speak to it,

00:15:39.360 | but like as a provider,

00:15:40.680 | we can just effectively give files to Netlify

00:15:43.400 | without the user even logging in,

00:15:44.800 | and they have a live website.

00:15:46.200 | And if they want to hold onto it,

00:15:47.880 | they can click a link and claim it

00:15:49.240 | to their Netlify account.

00:15:50.640 | But it basically is just this really magic experience.

00:15:53.800 | 'Cause when you come to Bolt, you say, "I want a website."

00:15:55.480 | Like my mom, 71 years old,

00:15:57.040 | made her first website, you know,

00:15:58.480 | on the internet two weeks ago, right?

00:16:00.400 | It was about her nursing days.

00:16:01.920 | And so it had made up--

00:16:03.040 | - Oh, that's fantastic, though.

00:16:04.520 | It wouldn't have been made.

00:16:05.800 | - 100%, 'cause even in, you know,

00:16:07.160 | when we had a lot of people building personal,

00:16:10.560 | like deeply personal stuff,

00:16:11.720 | like in the first week we launched this,

00:16:13.560 | the sales guy from the East Coast, you know,

00:16:15.720 | replied to a tweet of mine, and he said,

00:16:17.360 | "Thank you so much for building this."

00:16:18.760 | You know, to your team,

00:16:20.240 | his daughter has a medical condition.

00:16:21.920 | And so for her to travel,

00:16:23.080 | she has to like line up donors or something,

00:16:25.520 | you know, so ahead of time.

00:16:26.880 | And so he actually used Bolt to make a website to do that,

00:16:30.000 | to actually go and send it to folks

00:16:31.480 | in the region she was gonna travel to ahead of time.

00:16:33.880 | I was really touched by it.

00:16:35.120 | And I also thought like, why, you know,

00:16:36.240 | why didn't he use like Wix or Squarespace, right?

00:16:38.120 | I mean, this is a solved problem, quote unquote, right?

00:16:41.320 | And then when I thought, I actually used Squarespace

00:16:43.080 | for my, the wedding website for my wife and I,

00:16:45.880 | like back in 2021.

00:16:46.720 | So I'm familiar, you know,

00:16:48.200 | it was faster.

00:16:49.120 | I know how to code, but I was like, this is faster, right?

00:16:50.800 | And I thought back and I was like,

00:16:51.960 | there's a whole interface you have to learn how to use.

00:16:53.600 | And it's actually not that simple.

00:16:55.320 | There's like a million things

00:16:56.280 | you can configure in that thing.

00:16:57.440 | When you come to Bolt, there's a text box.

00:16:59.400 | You just say, I need a wedding website.

00:17:01.720 | Here's the date, here's where it is.

00:17:04.120 | And here's a photo of me and my wife.

00:17:06.200 | Put it somewhere relevant.

00:17:07.240 | It's actually the simplest way.

00:17:08.320 | And that's what, when my mom came, she said,

00:17:10.520 | I'm Pat Simons.

00:17:11.720 | I was a nurse in the '70s, you know,

00:17:14.200 | and like, here's the things I did.

00:17:16.120 | And a website came out.

00:17:18.160 | So coming back to why is this such a,

00:17:19.920 | I think, why are we seeing this sort of growth?

00:17:21.600 | It's, this is the simplest interface

00:17:23.720 | I think maybe ever created

00:17:25.080 | to actually build and deploy a website.

00:17:27.440 | 'Cause, and then that website my mom made,

00:17:29.520 | she's like, okay, this looks great.

00:17:31.000 | And there's one button.

00:17:32.480 | You just click it, deploy, and it's live.

00:17:34.080 | And you can buy a domain name, attach it to it.

00:17:36.920 | And you know, it's as simple as it gets.

00:17:38.240 | It's getting even simpler

00:17:39.240 | with some of the stuff we're working on.

00:17:40.320 | But anyways, so that's,

00:17:41.760 | it's been really interesting

00:17:44.320 | to see some of the usage like that.

00:17:45.840 | - I can offer my perspective.

00:17:47.280 | So I, you know, I probably should have disclosed

00:17:49.880 | a little bit that I'm a Slacklist investor.

00:17:52.800 | - Cancel the episode.

00:17:53.640 | - I know, I know.

00:17:54.480 | - Don't play it, no.

00:17:55.840 | Pause.

00:17:56.680 | - Eric actually reached out to show me Bolt

00:17:58.440 | before the launch.

00:17:59.280 | And we, you know, we talked a lot about like

00:18:01.560 | the framing of what we're gonna talk,

00:18:04.160 | how we marketed the thing,

00:18:06.000 | but also like what Bolt was going to need,

00:18:08.560 | like a whole sort of infrastructure.

00:18:10.360 | Netlify, I was a maintainer of the CLI,

00:18:12.280 | but I won't take claim for the anonymous upload.

00:18:14.840 | That's actually the origin story of Netlify.

00:18:16.240 | You can have Matt Billman talk about it,

00:18:17.800 | but he, like that was how Netlify started.

00:18:20.560 | You could drag and drop your zip file or folder

00:18:24.000 | from your desktop onto a website.

00:18:25.280 | It would have a live URL with no sign-in.

00:18:27.800 | And so that was the origin story of Netlify.

00:18:29.600 | And it just persisted to today.

00:18:31.360 | And it's just like, it's really nice, interesting

00:18:33.800 | that both Bolt and CognitionDevIn

00:18:36.520 | and a bunch of other sort of agent type startups,

00:18:38.360 | they all use Netlify to deploy because of this one feature.

00:18:42.240 | They don't really care about the other features of Netlify.

00:18:45.160 | But just because it's easy for computers to use

00:18:47.560 | and talk to, like if you build an interface

00:18:49.920 | for computers specifically that is easy for them to navigate,

00:18:53.000 | then they will be used in agents.

00:18:54.480 | And I think that's a learning

00:18:55.760 | that a lot of developer tools companies are having.

00:18:58.480 | That's my Bolt launch story and Netlify CLI stuff.

00:19:00.600 | And then I just wanted to come back

00:19:01.560 | to like the web containers things, right?

00:19:03.760 | Like I think you put a lot of weight on the technical modes.

00:19:06.920 | I think you also are just like very good at product.

00:19:09.680 | So you've like built a better agent than a lot of people,

00:19:12.680 | the rest of us, including myself,

00:19:14.520 | who have tried to build these things

00:19:15.720 | and we didn't get as far as you did.

00:19:17.360 | Don't shortchange yourself on products.

00:19:18.920 | But I think specifically on infra, on like the sandboxing,

00:19:22.000 | like this is a thing that people really want.

00:19:23.960 | Alessio has backed e2b, which we'll have on at some point

00:19:26.160 | talking about like the sort of the server full side,

00:19:28.280 | but yours is, you know, inside of the browser, serverless,

00:19:31.720 | that it doesn't cost you anything to serve one person

00:19:33.960 | versus a million people, it doesn't cost you anything.

00:19:36.160 | I think that's interesting.

00:19:37.440 | I think in theory, we should be able to like run tests

00:19:40.040 | 'cause you can run the full backend.

00:19:41.840 | Like you can run Git, you can run Node,

00:19:44.160 | you can run maybe Python someday.

00:19:47.040 | We talked about this.

00:19:48.120 | But ideally you should be able to have a full agentic loop,

00:19:51.080 | running code and seeing the errors, correcting code

00:19:53.320 | and just kind of self healing, right?

00:19:56.040 | Like, I mean, isn't that the dream?

00:19:57.800 | - Totally. - Yeah.

00:19:58.720 | - Totally.

00:19:59.560 | At least in bulk, we've got a good amount of that today.

00:20:02.120 | I mean, there's a lot more for us to do,

00:20:03.720 | but one of the nice things, 'cause like in web container,

00:20:07.160 | you know, there's a lot of kind of stuff you go Google,

00:20:08.960 | like, you know, turn Docker container into wasm.

00:20:11.480 | Like there's, you'll find a lot of stuff out there

00:20:13.720 | that will do that.

00:20:14.600 | The problem is it's very big, it's slow,

00:20:16.440 | and then that ruins the experience.

00:20:18.160 | And so what we ended up doing

00:20:19.200 | is just writing an operating system from scratch

00:20:21.040 | that was just purpose-built to, you know,

00:20:22.960 | run in a browser tab.

00:20:24.600 | And the reason being is, you know,

00:20:25.680 | these Docker to wasm things will give you an image

00:20:28.360 | that's like out 60 to 100 meg, you know, maybe more,

00:20:30.960 | you know, and our OS, you know, kind of clocks in,

00:20:34.080 | I think we're in like a, maybe a megabyte or less

00:20:37.280 | or something like that.

00:20:38.120 | I mean, it's, you know, really, really, you know,

00:20:40.560 | stripped down.

00:20:41.520 | - Basically the task involved, as I understand it,

00:20:43.920 | it's mapping every single Linux syscall

00:20:47.360 | to some kind of WebAssembly implementation.

00:20:49.680 | - But more or less.

00:20:50.520 | And then there's a lot of things actually,

00:20:52.240 | like when you're looking at a dev environment,

00:20:54.280 | there's a lot of things that you don't need

00:20:56.520 | that a traditional OS is gonna have, right?

00:20:58.800 | Like, you know, audio drivers,

00:21:00.720 | or you're like, there's just like,

00:21:01.560 | there's just tons of things, right?

00:21:03.400 | You can kind of toss them,

00:21:05.080 | or alternatively what you can do is you can actually be,

00:21:08.240 | the nice thing, and this kind of comes back

00:21:09.960 | to the origins of browsers,

00:21:11.920 | which is, at the beginning of the web in the late '90s,

00:21:15.400 | there was two very different kind of visions for the web

00:21:19.000 | where Alan Kay vehemently disagreed with the idea

00:21:21.640 | that it should be document-based,

00:21:22.840 | which is, you know, Tim Berners-Lee,

00:21:25.240 | and that's kind of what ended up winning

00:21:26.680 | was this document-based kind of browsing documents

00:21:28.760 | on the web thing.

00:21:29.800 | Alan Kay, he's got this like very famous quote

00:21:32.120 | where he said, "You know, you want web browsers

00:21:34.480 | "to be mini operating systems.

00:21:36.160 | "They should download little mini binaries

00:21:38.000 | "and execute with like a little mini

00:21:40.080 | "virtualized operating system in there."

00:21:42.160 | And what's kind of interesting about the history,

00:21:43.760 | not to geek out on this estimate,

00:21:45.080 | what's kind of interesting about the history

00:21:46.120 | is both of those folks ended up being right.

00:21:47.920 | Documents were actually the pragmatic way

00:21:50.560 | that the web became the most ubiquitous platform

00:21:53.080 | in the world, to the degree now

00:21:54.760 | that this is why WebAssembly has been invented,

00:21:57.320 | is that we need to do more low-level things in a browser,

00:21:59.720 | same thing with WebGPU, et cetera.

00:22:01.560 | And so all these APIs, you know,

00:22:03.400 | to build an operating system came to the browser,

00:22:05.480 | and that was actually the realization we had in 2017

00:22:07.880 | was, holy heck, like you can actually, you know,

00:22:10.560 | service workers, which were designed

00:22:12.480 | for allowing your app to work offline,

00:22:14.520 | that was the kind of the key one

00:22:15.880 | where it was like, wait a second,

00:22:17.360 | you can actually now run web servers within a browser.

00:22:20.960 | Like you can run a server that you open up, that's wild.

00:22:23.960 | - Like full Node.js.

00:22:25.080 | - Full Node.js, like that capability,

00:22:26.760 | like I can have a URL that's programmatically controlled

00:22:29.680 | by a web application itself, boom.

00:22:32.280 | Like the web can build the web, the primitive is there.

00:22:34.920 | Everyone at the time, like we talked to people

00:22:36.280 | that like worked on, you know, Chrome and V8,

00:22:38.280 | and they were like, oh, you know, like, I don't know.

00:22:41.200 | But it's one of those things,

00:22:42.040 | you just kind of have to go do it to find out.

00:22:43.680 | So we spent a couple of years, you know,

00:22:44.880 | working on it and got it to work.

00:22:47.000 | And back in 2021 is when we kind of put the first like data

00:22:50.400 | of WebContainer online, but-

00:22:52.680 | - In partnership with Google, right?

00:22:53.800 | Like Google actually had to help you

00:22:55.280 | get over the finish line with some stuff.

00:22:56.600 | - Yeah, 100%, because what, you know,

00:22:58.880 | over the years of when we were doing the R&D on the thing,

00:23:01.480 | kind of the biggest challenge,

00:23:03.000 | the two ways that you can kind of test

00:23:04.720 | how powerful and capable a platform are,

00:23:07.560 | the two types of applications are one, video games, right?

00:23:10.680 | 'Cause they're just very compute intensive,

00:23:12.520 | a lot of calculations that have to happen, right?

00:23:15.520 | The second one are IDEs,

00:23:17.080 | because you're talking about actually virtualizing

00:23:20.880 | the actual runtime environment you are in

00:23:23.040 | to actually build apps on top of it,

00:23:24.560 | which requires sophisticated capabilities,

00:23:27.800 | a lot of access to, you know,

00:23:29.920 | a good amount of compute power, right?

00:23:31.400 | To effectively, you know, building app in app sort of thing.

00:23:34.400 | So those are the stress tests.

00:23:35.840 | So if your platform is missing stuff,

00:23:38.080 | those are the things where you find out.

00:23:39.760 | Those are the people building games and IDEs,

00:23:42.120 | they're the ones filing bugs on operating system level stuff,

00:23:45.520 | and for us, browser level stuff.

00:23:46.720 | And so, yeah, what ended up happening

00:23:47.960 | is we were just hammering, you know,

00:23:49.240 | the Chromium bug tracker,

00:23:50.760 | and they're like, "Who are these guys?"

00:23:52.440 | - Yeah. (laughs)

00:23:54.280 | - And they were amazing, 'cause I mean,

00:23:56.200 | just making Chrome DevTools be able to debug,

00:23:58.360 | I mean, it wasn't originally built right

00:24:01.120 | for debugging an operating system, right?

00:24:04.040 | They've been phenomenal working with us

00:24:05.640 | and just kind of really pushing the limits.

00:24:07.480 | But it's a rising tide that's kind of lifted all boats,

00:24:09.800 | because now there's a lot of different types

00:24:11.800 | of applications that you can debug with Chrome DevTools

00:24:14.360 | that are running a browser that runs more reliably,

00:24:16.600 | because just the stress testing that we,

00:24:19.160 | and, you know, games that are coming to the web

00:24:21.360 | are kind of pushing as well, but.

00:24:23.040 | - That's awesome.

00:24:23.880 | About the testing, it seems like most,

00:24:26.320 | let's say, coding assistants from different kinds

00:24:28.480 | will need this loop of testing,

00:24:30.520 | and even I would add code review

00:24:32.480 | to some extent that you mentioned.

00:24:34.480 | - How is testing different from code review?

00:24:35.840 | - Code review could be, for example--

00:24:36.680 | - Just PR review?

00:24:37.760 | - PR review is like a code review that is done

00:24:40.200 | at the point of when you want to merge code branches.

00:24:43.920 | But I would say that code review, for example,

00:24:46.200 | checks best practices, maintainability, et cetera.

00:24:49.400 | - It's not just like CI, but it's more than CI.

00:24:51.680 | - And testing is more like checking functionality,

00:24:54.440 | et cetera, so it's different.

00:24:56.000 | We call, by the way, all of these together code integrity,

00:24:58.520 | but that's a different story.

00:24:59.360 | Just to go back to the testing and--

00:25:01.760 | - Still here? - Specifically.

00:25:02.840 | - Yeah, since the first slide.

00:25:07.080 | - Yeah, we're consistent.

00:25:08.200 | So if we go back to the testing,

00:25:10.400 | I think it's not surprising that for us,

00:25:12.400 | testing is important, and for Bolt,

00:25:13.720 | it's testing is important, but I want to shed some light

00:25:16.680 | on a different perspective of it.

00:25:19.040 | Let's think about autonomous driving.

00:25:20.760 | Those startups that are doing autonomous driving

00:25:22.600 | for highway and autonomous driving for the city,

00:25:27.120 | and I think we saw the autonomous of the highway

00:25:31.360 | much faster and reaching to level, I don't know,

00:25:35.200 | four or so, much faster than those in the city.

00:25:38.680 | Now, in both cases, you need testing,

00:25:41.200 | and quote-unquote testing, verifying validation

00:25:44.960 | that you're doing the right thing on the road

00:25:46.720 | and you're reading and et cetera,

00:25:48.720 | but it's probably so different in the city

00:25:51.720 | that it could be actually different technology,

00:25:53.840 | and I claim that we're seeing something similar here.

00:25:56.360 | So when you're building the next Wix,

00:25:58.480 | and if I was them, I was looking at you

00:26:00.880 | and being a bit scared.

00:26:02.480 | That's what you're disrupting, what you just said.

00:26:04.800 | Then basically, I would say that, for example,

00:26:07.160 | that UX/UI is freaking important,

00:26:10.160 | and because you're more aiming for the end user.

00:26:14.520 | In this case, maybe it's an end user

00:26:16.280 | that doesn't know how to develop.

00:26:17.760 | For developers, it's also important,

00:26:19.600 | but let alone those that do not know how to develop.

00:26:21.640 | They need a slick UI/UX, and I think that's one reason,

00:26:25.160 | for example, I think Cursor have really good technology.

00:26:27.960 | I don't know the underlying, what's under their hood,

00:26:30.160 | but at least what they're saying,

00:26:31.680 | but I think also their UX/UI is great.

00:26:34.000 | It's a lot because they did their own ID,

00:26:37.560 | while if you're aiming for the city AI,

00:26:41.680 | suddenly there's a lot of testing and code review technology

00:26:46.320 | that's not necessarily that important.

00:26:50.000 | For example, let's talk about integration tests.

00:26:53.160 | Probably a lot of what you're building in Bolt at the moment

00:26:57.240 | is isolated applications.

00:27:01.240 | Maybe the vision or the end game

00:27:04.640 | is maybe having one solution for everything.

00:27:07.760 | It could be that eventually the highway AI companies

00:27:12.480 | will go into the city and the other way around,

00:27:15.080 | but at the beginning, there is a difference,

00:27:17.400 | and integration tests are a good example.

00:27:19.240 | I guess they're a bit less important,

00:27:21.400 | and when you think about enterprise software,

00:27:23.640 | they're really important.

00:27:24.720 | So to recap, I think the idea of looping and verifying your test

00:27:29.440 | and verifying your code in different ways,

00:27:31.840 | testing or code review, et cetera,

00:27:33.720 | seems to be important in the highway AI and the city AI,

00:27:36.920 | but in different ways, and critical for the city,

00:27:41.040 | even more and more variety.

00:27:42.560 | Actually, I was looking to ask you,

00:27:46.080 | what kind of loops you guys are doing?

00:27:47.880 | For example, when I'm using Bolt and I'm enjoying it a lot,

00:27:50.720 | then I do see sometimes you're trying

00:27:53.880 | to catch errors and fix them.

00:27:55.480 | And also, I notice that you're breaking down tasks

00:27:58.520 | into smaller ones, et cetera, which

00:28:00.360 | is already a common notion from a year ago,

00:28:03.600 | but it seems like you're doing it really well.

00:28:05.560 | So if you're willing to share anything about that.

00:28:07.840 | Yeah, yeah, totally.

00:28:08.960 | I realize I never actually hit the punchline of what

00:28:11.280 | I was saying before.

00:28:12.280 | I mentioned the point about us writing an operating

00:28:14.120 | system from scratch, because what ended up

00:28:15.880 | being important about that is that, to your point,

00:28:18.600 | it's actually a very--

00:28:19.880 | compared to a-- if you're running a cursor on anyone's

00:28:23.560 | machine, you kind of don't know what

00:28:25.160 | you're dealing with with the OS you're running on.

00:28:26.820 | There could be an error happens.

00:28:28.280 | It could be like a million different things, right?

00:28:30.680 | There could be some config.

00:28:31.840 | It could be god knows what, right?

00:28:33.440 | The thing with WebConnector is because we wrote the entire

00:28:35.800 | thing from scratch.

00:28:36.600 | It's actually a unified image, basically.

00:28:39.640 | And we can instrument it at any level

00:28:42.280 | that we think is going to be useful, which

00:28:44.080 | is exactly what we did when we started building Bolt,

00:28:46.720 | is we instrumented stuff at the process level,

00:28:49.400 | at the runtime level, et cetera, et cetera, et cetera.

00:28:52.040 | Stuff that would just be not impossible to do on local,

00:28:54.960 | but to do that in a way that works across any operating

00:28:57.440 | system, whatever, I mean, would just be insanely difficult

00:29:01.640 | to do right and reliably.

00:29:03.360 | And that's what you saw when you've used Bolt,

00:29:05.860 | is that when an error actually will occur,

00:29:07.600 | whether it's in the build process

00:29:08.980 | or the actual web application itself is failing

00:29:11.120 | or anything kind of in between, you can actually

00:29:13.200 | capture those errors.

00:29:14.240 | And today, it's a very primitive way

00:29:16.280 | of how we've implemented it, largely because the product just

00:29:18.920 | didn't exist 90 days ago.

00:29:20.000 | So we got some work out of us, and we got to hire some more

00:29:23.360 | a little bit.

00:29:23.920 | But basically, we present it.

00:29:25.480 | We say, hey, here's kind of the things that went wrong.

00:29:28.200 | There's a Fix It button and then an Ignore button.

00:29:30.360 | And then you can just hit Fix It,

00:29:31.720 | and then we take all that telemetry through our agent

00:29:35.680 | and say, here's the state of the application.

00:29:37.800 | Here's the errors that we got from Node.js or the browser

00:29:41.360 | or whatever, and da, da, da, da.

00:29:42.800 | And it can take a crack at actually solving it.

00:29:44.760 | And it's actually pretty darn good at being able to do that.

00:29:47.560 | That's kind of been a--

00:29:48.600 | closing the loop and having it be a reliable kind of base

00:29:52.360 | has seemed to be a pretty big upgrade over doing stuff

00:29:56.000 | locally, just because I think that's

00:29:57.920 | a pretty key ingredient of it.

00:29:59.480 | And yeah, I think breaking things down into smaller tasks,

00:30:01.920 | that's kind of a key part of our agent.

00:30:03.640 | And I think Claude did a really good job with artifacts.

00:30:06.560 | And I think us and kind of everyone else

00:30:08.400 | has kind of taken their approach of actually breaking out

00:30:11.640 | certain tasks in a certain order into a concrete way.

00:30:14.960 | And so actually, the core of Bolt.io,

00:30:16.840 | we actually made open source.

00:30:18.360 | So you can actually go and check out the system prompts

00:30:20.960 | and et cetera, and you can run it locally

00:30:23.040 | and whatever have you.

00:30:24.160 | So anyone that's interested in this stuff,

00:30:25.900 | I'd highly recommend taking a look at.

00:30:28.440 | There's not a lot of stuff that's

00:30:30.360 | open source in this realm.

00:30:31.520 | That was one of the fun things that we thought

00:30:33.720 | would be cool to do, and people seem to like it.

00:30:36.380 | I mean, there's a lot of forks and people adding

00:30:38.380 | different models and stuff, so it's been cool to see.

00:30:40.560 | Yeah, I'm happy to add--

00:30:42.120 | I added real-time voice for my opening Dev Day demo,

00:30:45.440 | and it was really fun to hack with.

00:30:46.960 | So thank you for doing that.

00:30:48.200 | You didn't have to.

00:30:48.960 | I'm going to steal your code back.

00:30:50.400 | It's our fault, because I want that note.

00:30:52.360 | It's funny, because I built on top of the fork of Bolt.io

00:30:54.920 | that already has the multi-LLM thing.

00:30:56.880 | And so you just told me you're going to merge that in,

00:30:58.320 | so then you're going to merge two layers of forks

00:31:00.360 | down into this thing.

00:31:01.680 | So it'll be fun.

00:31:02.640 | Heck yeah.

00:31:03.560 | Just to touch on the environment,

00:31:05.200 | Eramar, you maybe go into the most complicated environments

00:31:08.280 | that even the people that work there don't know how to run.

00:31:10.440 | How much of an impact does that have on your performance?

00:31:15.040 | Like, you know, is most of the work you're doing actually

00:31:17.280 | figuring out environment and, like, the libraries?

00:31:19.120 | Because I'm sure they're using outdated version of languages.

00:31:22.040 | They're using outdated libraries.

00:31:23.600 | They're using forks that have not been on the public internet

00:31:26.440 | before.

00:31:27.120 | How much of the work that you're doing

00:31:28.720 | is, like, there versus, like, at the LLM level?

00:31:31.440 | One of the reasons I was asking about, you know,

00:31:34.880 | what are the steps to break things down,

00:31:36.960 | because it really matters, like, what's the tech stack,

00:31:40.760 | how complicated the software is.

00:31:42.600 | It's hard to figure it out when you're

00:31:44.200 | dealing with the real world, any environment of enterprise.

00:31:48.520 | It's a city, what I'm--

00:31:49.720 | like, while maybe sometimes, like, I

00:31:52.640 | think you do enable, like, and bolt, like, to install stuff,

00:31:56.240 | but it's quite a, like, controlled environment.

00:31:58.800 | And that's a good thing to do, because then you narrow it down

00:32:02.280 | and it's easier to make things work.

00:32:05.000 | So definitely there are two dimensions.

00:32:06.840 | I think, actually, spaces.

00:32:08.720 | One is the fact, just, like, installing our software

00:32:12.320 | without yet, like, making it work, just installing it.

00:32:16.280 | Because we work with enterprise and Fortune 500, et cetera,

00:32:19.440 | many of them want on-prem solution.

00:32:21.440 | So--

00:32:22.000 | You have how many deployment options?

00:32:23.560 | Basically, we did a metrics, let's say, 96 options.

00:32:28.760 | Because, you know, there are different dimensions.

00:32:31.000 | Like, for example, one dimension,

00:32:32.840 | we connect to your code management system,

00:32:35.800 | to your Git, so you're having, like, GitHub, GitLab--

00:32:39.640 | Subversion.

00:32:40.760 | Is it, like, on cloud, or deployed on-prem?

00:32:44.320 | Just an example.

00:32:45.320 | Which model agree to use its APIs, or ours?

00:32:48.320 | Like, we have our own models--

00:32:49.560 | Is it TestGPT, or is it--

00:32:50.720 | Yeah, we started with TestGPT.

00:32:52.480 | It was a huge mistake, the name.

00:32:54.920 | It was cool at the beginning, but I

00:32:56.480 | don't think it's a good idea to name a model

00:32:58.440 | after someone else's model.

00:33:00.120 | Anyway, that's my opinion, so we--

00:33:02.080 | I'm interested in these learnings, like, things

00:33:04.080 | that you change your mind on.

00:33:05.560 | Eventually, when you're building a company,

00:33:07.320 | you're building a brand.

00:33:08.280 | And you want to create your own brand.

00:33:10.920 | By the way, when I thought about Bolt.new,

00:33:12.760 | I also thought about if it's not a problem.

00:33:14.600 | Because when I think about Bolt, I

00:33:16.040 | do think about, like, a couple of companies

00:33:17.880 | I already called as well.

00:33:19.200 | Curse companies.

00:33:21.280 | You could call it Codium, just to--

00:33:23.960 | OK, thank you.

00:33:25.200 | Touche, touche.

00:33:27.000 | Yeah, you've got to imagine, the board

00:33:28.840 | meeting before we launched Bolt, one of our investors--

00:33:31.200 | you can imagine, they were like, are you sure?

00:33:33.360 | Because from the investment side,

00:33:34.800 | it was kind of a famous, very notorious Bolt.

00:33:37.640 | And they're like, are you sure you want to go with that name?

00:33:40.240 | Oh, yeah, yeah, absolutely.

00:33:42.200 | At this point, we have actually four models.

00:33:44.040 | There is a model for autocomplete.

00:33:46.320 | There is a model for the chat.

00:33:47.600 | There is a model dedicated for more for code review.

00:33:50.200 | And there is a model that is for code embedding.

00:33:53.280 | Actually, you might notice that there isn't a good code

00:33:57.120 | embedding model out there.

00:33:59.120 | Can you name one that would dedicate it for code?

00:34:03.080 | Yeah, I can share more than that.

00:34:04.480 | There's code indexing.

00:34:05.440 | And then you can do sort of like the hide for code.

00:34:08.680 | And then you can embed the descriptions of the code.

00:34:12.000 | Yeah, but you do see a lot of type

00:34:14.840 | of models that are dedicated for embedding

00:34:17.240 | and for different spaces, different fields, et cetera.

00:34:21.240 | And I'm not aware.

00:34:23.360 | And I know that if you go to the bedrock,

00:34:26.040 | try to find there is a few code embedding models.

00:34:28.280 | But none of them are specialized for code.

00:34:31.040 | Is there a benchmark that you would

00:34:32.560 | tell us to pay attention to?

00:34:33.960 | Yeah, so it's coming.

00:34:35.960 | Wait for that.

00:34:36.500 | Anyway, we have our models.

00:34:38.600 | And just to go back to the '96 option of deployment--

00:34:43.440 | Yeah, sorry, I can't hear you.

00:34:44.720 | Yeah, so I'm closing the brackets for us.

00:34:47.160 | So one is dimensional, like what Git deployment you have.

00:34:51.600 | What models do you agree to use?

00:34:53.920 | Dotter could be like if it's air-gapped completely,

00:34:56.880 | or you want VPC.

00:34:57.960 | And then you have Azure GCP and AWS, which is different.

00:35:02.560 | Do you use Kubernetes or do not?

00:35:04.520 | Because we want to exploit that.

00:35:06.480 | There are companies that do not do that, et cetera.

00:35:09.000 | I guess you know what I mean.

00:35:10.240 | So that's one thing.

00:35:11.040 | And considering that we are dealing

00:35:12.560 | with one of all four enterprise, we needed to deal with that.

00:35:17.000 | So you asked me about how complicated

00:35:19.080 | it is to solve that complex code.

00:35:20.640 | I said, it's just the deployment part.

00:35:22.680 | And then now to the software, we see

00:35:26.680 | a lot of different challenges.

00:35:28.760 | For example, some companies, they

00:35:31.680 | did actually a good job to build a lot of microservices.

00:35:36.200 | Let's not get to if it's good or not.

00:35:38.320 | But let's for a second assume that it is a good thing.

00:35:41.360 | A lot of microservices, each one of them has their own repo.

00:35:44.040 | And now you have tens of thousands of repos.

00:35:47.040 | And you, as a developer, want to develop something.

00:35:49.400 | And I remember me coming to a corporate for the first time.

00:35:53.440 | And I don't know where to look at, where to find things.

00:35:56.880 | So just doing a good indexing for that is a challenge.

00:36:02.200 | And moreover, the regular indexing,

00:36:04.640 | the one that you can find-- well, we wrote a few blogs

00:36:06.840 | on that, by the way.

00:36:07.640 | We also have some open source, different than yours,

00:36:10.320 | but actually three and growing.

00:36:12.320 | Then it doesn't work.

00:36:13.480 | You need to let the tech leads in the companies

00:36:15.480 | influence your indexing.

00:36:16.960 | For example, Mark with different repos with different colors.

00:36:22.600 | This is a high-quality repo.

00:36:23.920 | This is a lower-quality repo.

00:36:25.680 | This is a repo that we want to deprecate.

00:36:27.720 | This is a repo we want to grow, et cetera.

00:36:30.520 | And let that be part of your indexing.

00:36:33.320 | And only then things actually work for enterprise.

00:36:36.720 | And they don't get to a fatige of, oh, this is awesome.

00:36:39.720 | Oh, but I'm starting-- it's annoying me.

00:36:42.680 | I think Copilot is an amazing tool.

00:36:44.480 | But I'm quoting others, meaning GitHub Copilot,

00:36:47.640 | that they see not so good retention of GitHub Copilot

00:36:51.400 | in enterprise.

00:36:52.600 | Ooh, spicy.

00:36:53.560 | I saw snapshots of people--

00:36:55.920 | and we have customers that are Copilot users as well.

00:36:59.240 | And also, I saw research--

00:37:02.200 | some of them is public, by the way--

00:37:03.840 | between 38% to 50% retention for users using Copilot

00:37:08.600 | in enterprise.

00:37:09.960 | So it's not so good, right?

00:37:11.560 | By the way, I don't think it's that bad.

00:37:13.320 | But it's not so good.

00:37:14.680 | So I think that's a reason.

00:37:15.920 | Because, yeah, it helps you auto-complete.

00:37:18.000 | But then, and especially if you're

00:37:20.080 | working on your repo alone, but if it

00:37:22.840 | needs that context of remote repos that are code-based,

00:37:26.680 | that's hard.

00:37:27.240 | So to make things work, there's a lot of work on that,

00:37:30.000 | like giving the controllability for the tech leads,

00:37:34.360 | for the developer platform, or developer experience

00:37:38.000 | department in the organization to influence

00:37:40.600 | how things are working.

00:37:42.480 | A short example, because already on,

00:37:44.680 | if you have really old legacy code, probably some of it

00:37:48.360 | is not so good anymore.

00:37:50.320 | If you just fine-tune on these code base,

00:37:53.080 | then there is a bias to repeat those mistakes,

00:37:56.080 | or old practices, et cetera.

00:37:57.640 | So you need, for example, as I mentioned,

00:38:01.680 | to influence that.

00:38:02.440 | For example, in Coda, you can have a markdown

00:38:05.040 | of best practices by the tech leads.

00:38:07.400 | And Coda will include that, and relate to that,

00:38:10.400 | and will not offer suggestions that are not

00:38:13.080 | according to the best practices, just as an example.

00:38:16.000 | So that's just a short list of things

00:38:18.720 | that you need to do in order to deal with, like you mentioned,

00:38:21.880 | the 100.1 to 100.2 version of software.

00:38:28.200 | I just want to say, what you're doing is extremely impressive,

00:38:30.920 | because it's very difficult. I mean,

00:38:32.480 | the business of StackPulse, kind of before bulk came online,

00:38:35.000 | we sold a version of our IDE that went on-prem.

00:38:38.160 | So I understand what you're saying

00:38:39.680 | about the difficulty of getting stuff just working on-prem.

00:38:42.920 | Holy heck.

00:38:43.640 | I mean, that is extremely hard.

00:38:45.760 | I guess the question I have for you is,

00:38:47.400 | I mean, we were just doing that with Kubernetes-based stuff.

00:38:50.040 | But the spread of Fortune 500 companies

00:38:51.760 | that you're working with, how are they

00:38:52.880 | doing the inference for this?

00:38:54.080 | Are you kind of plugging into Azure's OpenAI stuff,

00:38:58.040 | and AWS's Bedrock, you know, Cloud stuff?

00:39:00.920 | Or are they just running stuff on GPUs?

00:39:02.920 | Like, what is that?

00:39:03.920 | How are these folks approaching that?

00:39:05.520 | Because, man, from what we saw on the enterprise side,

00:39:07.960 | I mean, I got to imagine that that's a huge challenge.

00:39:10.920 | Everything you said, and more.

00:39:13.760 | For example, someone could be in a--

00:39:16.720 | I don't think any of these is bad.

00:39:18.960 | They made their decision.

00:39:20.280 | For example, some people, they're,

00:39:22.760 | I want only AWS and VPC on AWS, no matter what.

00:39:27.320 | And then some of them, there is a subset, I would say,

00:39:30.680 | I'm willing to take models only from Bedrock and not ours.

00:39:35.640 | And we have a problem, because there is no good code embedding

00:39:38.880 | model on Bedrock.

00:39:40.200 | And that's part of what we're doing now with AWS to solve that.

00:39:43.560 | We solve it in a different way.

00:39:44.560 | But if you are willing to run on AWS VPC,

00:39:47.400 | but run your run models on GPUs or Inferentia,

00:39:52.520 | like the new version, the more coming out,

00:39:54.800 | then our models can run on that.

00:39:56.400 | But everything you said is right.

00:39:58.040 | Like, we see on-prem deployment where

00:40:01.480 | they have their own GPUs.

00:40:03.120 | We see Azure where they're using OpenAI Azure.

00:40:06.680 | We see cases where you're running on GCP,

00:40:09.040 | and they want OpenAI, like this cross case.

00:40:13.640 | Although there is Gemini, or even Sonnet, I think,

00:40:16.200 | is available on GCP, just an example.

00:40:18.320 | So all the options, that's part of the challenge.

00:40:22.120 | I admit that we thought about it,

00:40:23.520 | but it was even more complicated.

00:40:25.160 | And it took us a few months to actually,

00:40:27.600 | that metrics that I mentioned, to start clicking

00:40:30.320 | each one of the blocks there.

00:40:32.280 | A few months is impressive.

00:40:33.400 | I mean, honestly, it just, that's okay.

00:40:35.800 | And every one of these enterprises is,

00:40:37.680 | their networking is different, just everything's different.

00:40:40.160 | Every single one is different.

00:40:41.160 | - As you understand.

00:40:42.560 | - Yeah, so that just cannot be understated.

00:40:44.840 | Like, that it is, that's extremely impressive.

00:40:48.160 | Hats off.

00:40:49.000 | - It could be, by the way, like, for example,

00:40:50.480 | oh, we're only AWS, but our GitHub enterprise is on-prem.

00:40:55.480 | Oh, we forgot.

00:40:56.920 | So we need, like, a private link, or whatever.

00:40:58.760 | Every time, like, it's not.

00:41:00.800 | And you do need to think about it

00:41:02.000 | if you want to work with enterprise.

00:41:04.160 | And it's important.

00:41:05.000 | - Yeah, I understand.

00:41:05.840 | Like, I respect their point of view.

00:41:08.240 | - And this primarily impacts your architecture,

00:41:10.400 | your tech choices.

00:41:11.360 | Like, you have to, you can't choose some vendors because--

00:41:13.920 | - Yeah, definitely.

00:41:14.760 | To be frank, it makes us hard for a startup

00:41:17.560 | because it means that we want everyone

00:41:20.000 | to enjoy all the variety of models.

00:41:22.640 | By the way, it was hard for us with our technology.

00:41:25.320 | I want to open a bracket, like a window.

00:41:28.680 | I guess you're familiar with our Alpha Codium,

00:41:31.120 | which is an open source--

00:41:32.200 | - We got to go over that.

00:41:33.200 | - Yeah, so I'll do that quickly.

00:41:34.960 | - A pin in that, yeah.

00:41:36.160 | - Actually, we didn't have it in the last episode, so okay.

00:41:39.880 | - Okay, we'll come back to that later,

00:41:40.840 | but let's talk about--

00:41:41.920 | - Yeah, so just, like, shortly,

00:41:43.400 | and then we can double-click on Alpha Codium.

00:41:45.320 | But Alpha Codium is an open source tool.

00:41:47.080 | You can go and try it,

00:41:48.120 | and it lets you compete on CodeForce.

00:41:50.800 | This is a website and a competition.

00:41:53.120 | And actually reach a master level, like 95%,

00:41:56.760 | with a click of a button.

00:41:57.720 | You don't need to do anything.

00:41:59.160 | And part of what we did there is taking a problem

00:42:03.360 | and breaking it to different, like, smaller blocks.

00:42:05.920 | And then the models are doing a much better job,

00:42:09.600 | like we all know it by now,

00:42:11.040 | that taking small tasks and solving them.

00:42:13.600 | By the way, even O1,

00:42:15.680 | which is supposed to be able to do system two thinking,

00:42:18.920 | like Greg from OpenAI, like, hinted,

00:42:22.320 | is doing better on these kind of problems.

00:42:24.520 | But still, it's very useful to break it down for O1,

00:42:27.960 | despite O1 being able to think by itself.

00:42:31.520 | And that's what we presented, like, just a month ago.

00:42:33.560 | OpenAI released it now.

00:42:35.600 | They are doing 93 percentile with O1 IOI left,

00:42:40.600 | and International Olympiad of Formation.

00:42:44.920 | Sorry, I forgot.

00:42:45.760 | Exactly.

00:42:46.600 | I told you I forgot.

00:42:47.600 | And we took their O1 preview with Alpha Codium

00:42:51.400 | and did better.

00:42:52.400 | Like, it just shows, like,

00:42:53.800 | and there is a big difference between the preview

00:42:55.720 | and the IOI.

00:42:57.440 | It shows, like, that these models

00:42:58.920 | are not still system two thinkers,

00:43:01.040 | and there is a big difference.

00:43:02.320 | Yeah, maybe they're not complete system two.

00:43:04.320 | Yeah, they need some guidance.

00:43:05.280 | I call them system 1.5.

00:43:07.080 | We can dive into that.

00:43:08.400 | I thought about it.

00:43:09.240 | Like, you know, I care about this philosophy stuff,

00:43:11.560 | and I think, like, we didn't see it

00:43:13.560 | even close to a system two thinking.

00:43:15.160 | I can elaborate later.

00:43:16.280 | But closing the bracket, it's like,

00:43:17.520 | we take Alpha Codium, and as our principle of thinking,

00:43:21.960 | we take tasks and break them down to smaller tasks.

00:43:25.800 | And then we want to exploit the best model to solve them.

00:43:28.800 | So I want to enable anyone to enjoy O1 and Sonnet

00:43:33.800 | and Gemini 1.5, et cetera.

00:43:36.520 | But at the same time,

00:43:37.920 | I need to develop my own models as well,

00:43:39.840 | because some of the Fortune 500

00:43:42.280 | want to have all air-gapped or whatever.

00:43:45.320 | So that's a challenge.

00:43:47.000 | Now, you need to support so many models.

00:43:49.920 | And to some extent, I would say that the flow engineering,

00:43:53.840 | the breaking down to different blocks

00:43:57.320 | is a necessity for us.

00:43:58.520 | Why?

00:43:59.360 | Because when you take a big block, a big problem,

00:44:02.680 | you need a very different prompt

00:44:04.720 | for each one of the models to actually work.

00:44:07.280 | But when you take, like, a big problem

00:44:09.640 | and break it into small tasks,

00:44:11.920 | we can talk how we do that.

00:44:13.640 | Then the prompt matters less.

00:44:16.120 | What I want to say, like, all this, like,

00:44:18.720 | as a startup trying to do different deployment,

00:44:21.400 | getting all the juice that you can get from models,

00:44:23.680 | et cetera, is a big problem.

00:44:25.440 | And one need to think about it.

00:44:27.200 | And one of our mitigation is that process

00:44:30.520 | of taking tasks and breaking them down.

00:44:32.360 | That's why I'm really interested to know

00:44:34.120 | how you guys are doing it.

00:44:35.120 | And part of what we do is also open source, so you can see.

00:44:38.160 | - There's a lot in there.

00:44:39.160 | But yeah, flow over prompt,

00:44:40.480 | I do believe that that does make sense.

00:44:43.360 | I feel like there's a lot that both of you

00:44:45.520 | can sort of exchange notes on breaking down problems.

00:44:48.560 | And I just want you guys to just go for it.

00:44:50.840 | Whatever, like, this is fun to watch.

00:44:52.800 | - Yeah, I mean, what's super interesting is

00:44:55.680 | the context you're working in is,

00:44:57.920 | 'cause for us too with Bolt,

00:44:59.840 | we've started thinking,

00:45:00.800 | because our kind of existing business line

00:45:03.360 | was going behind the firewall, right?

00:45:05.600 | We were like, how do we do this?

00:45:07.280 | Adding the inference aspect on, we're like, okay, how does,

00:45:09.840 | 'cause I mean, there's not a lot of prior art, right?

00:45:11.520 | I mean, like, this is all new, you know, this is all new.

00:45:13.840 | So I definitely am going to have a lot of questions for you.

00:45:16.040 | - I'm here.

00:45:17.080 | We're very open, by the way.

00:45:18.000 | We have a paper on it, a blog, like, whatever.

00:45:20.760 | - The Alphacodeum, GitHub,

00:45:22.040 | and we'll put all this in the show notes.

00:45:23.800 | - Yeah, and even the new results of '01, we published it.

00:45:28.160 | - I love that.

00:45:29.000 | I also just, I think spiritually,

00:45:30.360 | I like your approach of being transparent and, you know,

00:45:34.600 | 'cause I think there's a lot of hype-ium around AI stuff.

00:45:38.320 | And a lot of it is, it's just like,

00:45:40.080 | you have these companies

00:45:40.920 | that are just kind of keep their stuff closed source

00:45:42.080 | and then just max hype it,

00:45:43.760 | but then it's kind of nothing.

00:45:44.920 | And I think it kind of gives a bad rep

00:45:46.680 | to the incredible stuff that's actually happening here.

00:45:49.120 | And so I think it's stuff like what you're doing where,

00:45:51.200 | I mean, true merit and you're cracking open actual code

00:45:54.200 | for others to learn from and use, you know,

00:45:56.040 | that strikes me as the right approach.

00:45:58.240 | And it's great to hear that you're making such incredible.

00:46:01.000 | - I have something to share about open source.

00:46:02.400 | Like, most of our tools are,

00:46:04.240 | we have like an open source version

00:46:05.800 | and then like a premium pro version,

00:46:07.760 | but it's not an easy decision to do that.

00:46:09.800 | I actually wanted to ask you about your strategy,

00:46:11.440 | but I think like, in your case,

00:46:12.800 | there is, in my opinion, relatively a good strategy

00:46:15.840 | where a lot of parts of open source,

00:46:18.320 | but then you have the deployment and the environment,

00:46:20.720 | which is not, right, if I get it correctly.

00:46:22.640 | And then there's a clear, like,

00:46:24.280 | almost hugging face, like model, like,

00:46:25.920 | "Yeah, you can do that,

00:46:27.080 | but why should you try to deploy it yourself,

00:46:29.680 | deploy it with us?"

00:46:30.880 | But in our case, and I'm not sure you're not going to hit

00:46:33.360 | also some competitors, and I guess you are.

00:46:36.480 | I wanted to ask you, for example, on some of them.

00:46:38.600 | In our case, like one day,

00:46:40.200 | we looked on one of our competitors

00:46:42.040 | that is doing code review.

00:46:43.600 | We are a platform, we have like the code review,

00:46:45.520 | the testing, et cetera, spread over the ID to get,

00:46:48.360 | and in each agent, we have a few startups

00:46:50.960 | or a big, big, like incumbents that are doing only that.

00:46:54.960 | So we noticed one of our competitors

00:46:56.920 | having not only a very similar UI of our open source,

00:47:01.760 | but actually even our typo.

00:47:03.400 | (laughing)

00:47:05.000 | And you know, you sit there and you kind of like,

00:47:07.640 | "Yeah, we're not that good.

00:47:09.480 | Like, we don't use enough, like, grammarly or whatever."

00:47:12.040 | And we had like a couple of these and we saw it there.

00:47:16.560 | And then it's a challenge.

00:47:18.640 | Like, and I want to ask you, like, you know,

00:47:20.760 | Bald is doing so well, and then you open-sourced it.

00:47:23.200 | So I think I know what that, my answer was,

00:47:25.400 | I gave it before, but still interesting

00:47:27.600 | to hear what you think.

00:47:29.440 | - Geohot said back, I think it was when he was,

00:47:33.440 | I don't know what he was up to at this exact moment,

00:47:35.120 | but yeah, I think when on comma AI,

00:47:36.400 | you know, all that stuff's open source.

00:47:37.640 | And someone had asked him like, "Why is this open source?"

00:47:40.120 | And he's like, "If you're not actually confident

00:47:42.080 | that you can go and crush it and build the best thing,

00:47:45.720 | then yeah, you should probably

00:47:46.560 | keep your stuff closed source."

00:47:47.920 | Right?

00:47:48.760 | Like, he said something akin to that,

00:47:49.600 | I'm probably kind of butchering it,

00:47:50.680 | but I thought that was a really,

00:47:51.520 | it was kind of a really good point.

00:47:52.480 | And that's not to say that you should just

00:47:53.440 | open source everything, 'cause for obvious reasons,

00:47:55.360 | there's like kind of strategic things

00:47:56.880 | you have to kind of take in mind.

00:47:57.920 | But I actually think a pretty liberal approach,

00:47:59.360 | as liberal as you kind of can be,

00:48:01.080 | it can really make a lot of sense.

00:48:02.080 | 'Cause I mean, that is so validating

00:48:03.200 | that like one of your competitors is taking your stuff

00:48:05.920 | and they're like, "Yeah, like, let's just, you know,

00:48:08.160 | kind of tweak the styles."

00:48:09.480 | And you know, I mean, clearly, right?

00:48:11.560 | I think it's kind of healthy 'cause it keeps,

00:48:13.160 | I'm sure back at HQ that day when you saw that,

00:48:15.240 | you're like, "Oh, all right.

00:48:16.360 | Well, let's, we gotta, we have to grind even harder, right?

00:48:19.480 | To make sure we stay ahead."

00:48:20.600 | And so I think it's actually a very useful,

00:48:21.960 | motivating thing for the teams.

00:48:23.680 | 'Cause you might feel this period of comfort.

00:48:26.240 | I think a lot of companies will have this period of comfort

00:48:28.000 | where, you know, they're not feeling the competition,

00:48:30.480 | then one day they get disrupted.

00:48:31.800 | So kind of putting stuff out there

00:48:33.520 | and letting people push it,

00:48:35.320 | forces you to face reality soon, right?

00:48:37.640 | And actually feel that incrementally

00:48:40.600 | so you can kind of adjust course.

00:48:41.880 | And that's, for us, the open-source version of Bolt

00:48:44.520 | has had a lot of features.

00:48:45.800 | People have been begging us for like,

00:48:47.280 | persisting chat messages and checkpoints and stuff.

00:48:49.720 | Within the first week,

00:48:50.560 | that stuff was landed in the open-source versions.

00:48:52.400 | And they're like, "Why can't you ship this?

00:48:54.720 | It's in the open," so people have forked it.

00:48:56.600 | And we're like, "We're trying to keep our servers

00:48:59.680 | and GPUs online."

00:49:00.520 | Like we're, you know, like we're just,

00:49:02.120 | but it's been great 'cause it's, you know,

00:49:03.880 | like the folks in the community did a great job,

00:49:06.120 | kept us on our toes.

00:49:07.120 | And we've got to know most of these folks too,

00:49:10.120 | at this point, that have been building these things.

00:49:12.480 | And so it actually was very instructive.

00:49:14.040 | Like, okay, well, if we're gonna go kind of land this,

00:49:16.120 | there's some UX patterns we can kind of look at

00:49:17.800 | and the code is open-source to this stuff.

00:49:19.680 | You know, what's great about these, what's not.

00:49:21.000 | So anyway, so I think that's,

00:49:22.920 | yeah, anyways, NetNet, I think it's awesome.

00:49:24.920 | I think from a competitive point of view for us,

00:49:26.880 | for us, I think in particular, what's interesting

00:49:28.840 | is the kind of the core technology of web container going.

00:49:33.600 | And I think that right now,

00:49:36.040 | there's really nothing that's kind of on par with that.

00:49:38.560 | And we also like, we have a business of,

00:49:40.320 | 'cause, you know, web container runs in your browser,

00:49:42.080 | but to make it work, you have to install stuff from NPM.

00:49:45.000 | You have to like make cores bypass requests,

00:49:46.920 | like connected databases,

00:49:48.320 | which all require server-side, you know,

00:49:50.200 | proxying or acceleration.

00:49:51.760 | And so we actually sell web container as a service.

00:49:54.560 | The original, one of the core reasons we open-sourced

00:49:57.000 | kind of the core components of Bolt when we launched

00:49:59.480 | was that, you know, we think that

00:50:02.080 | there's gonna be a lot more of these AI,

00:50:04.600 | in-your-browser AI co-gen experiences,

00:50:07.120 | kind of like what Anthropic did with Artifacts and Clod.

00:50:10.000 | - By the way, Artifacts uses web containers, not yet.

00:50:13.200 | - No, yeah. - Should I strike that?

00:50:14.880 | - I think that they've got their own thing at the moment,

00:50:16.760 | but there's been a lot of interest in web containers

00:50:18.800 | from folks doing things in that sort of realm

00:50:21.320 | and in the AI labs and startups and everything in between,

00:50:24.960 | you know, and so I think, no, there'll be,

00:50:26.440 | I imagine a good, you know, over the coming months,

00:50:28.440 | there'll be lots of things to being announced

00:50:30.560 | to folks kind of adopting it.

00:50:31.640 | But yeah, I think effectively.

00:50:34.040 | - Okay, I'll say this.

00:50:34.880 | If you're a large model lab and you want, you know,

00:50:37.520 | to build, you know, sandbox environments

00:50:39.320 | inside of your chat app, you should call Eric.

00:50:41.360 | - But wait, wait, wait, wait, wait, wait, wait.

00:50:43.520 | Like, I have a question about that.

00:50:45.640 | - What? - Yeah.

00:50:46.480 | - Like, I think like OpenAI, you know,

00:50:48.640 | they felt that people are not using their model

00:50:52.160 | as they would want to, so they built ChatGPT,

00:50:56.760 | but I would say that ChatGPT now defines OpenAI.

00:51:00.880 | I know they're doing a lot of business from their APIs,

00:51:04.880 | but still, is this like how you think?

00:51:06.520 | Like, isn't Bolt.new your business now?

00:51:10.680 | Like, why don't you like focus on that instead of the--

00:51:14.240 | - What's your advice as a founder?

00:51:15.560 | - Well, you're right.

00:51:17.360 | And so going into it, we, candidly, we were like,

00:51:20.280 | okay, Bolt.new, this thing is super cool.

00:51:22.240 | I'm like, we think people are stoked.

00:51:23.800 | Well, we think people will be stoked,

00:51:25.000 | but we were like, maybe that's allowed, you know,

00:51:26.480 | at best case scenario, after month one,

00:51:29.120 | we'd be mind blown if we added a couple hundred K of error

00:51:31.600 | or something, you know?

00:51:32.760 | And we were like, but we think there's probably gonna be,

00:51:35.040 | you know, an immediate huge business,

00:51:36.760 | 'cause there was some early poll on, you know,

00:51:40.000 | folks wanting to put WebContainer

00:51:41.760 | into their product offerings, you know,

00:51:43.080 | kind of similar to what Bolt is doing or whatever.

00:51:45.800 | We were actually prepared for the inverse outcome here,

00:51:48.920 | but I mean, well, I guess we've seen poll on both,

00:51:50.800 | but I mean, what's happened with Bolt,

00:51:52.800 | and you're right, it's actually the same strategy

00:51:54.360 | as like OpenAI or Anthropic, where, you know,

00:51:57.120 | we have a, our chat GPT to OpenAI's APIs is,

00:52:01.320 | you know, Bolt to WebContainer.

00:52:03.360 | And so that's, we've kind of taken that same approach,

00:52:05.560 | and we're seeing, I guess, kind of some of the similar

00:52:07.880 | results, except our, right now the revenue side

00:52:10.640 | is extremely lopsided to Bolt, right?

00:52:14.880 | - I think like, if you ask me what's my advice,

00:52:16.840 | I think you have three options.

00:52:18.400 | One is to focus on Bolt.

00:52:20.840 | The other is to focus on the WebContainer.

00:52:23.080 | The third is to raise $1 billion or whatever,

00:52:27.320 | and do them both.

00:52:28.360 | I'm serious, like, I think like,

00:52:29.960 | otherwise you need to choose.

00:52:32.280 | And if you raise enough money, like,

00:52:34.560 | and I think it's big bucks,

00:52:35.720 | because you're going to be chased by competitors,

00:52:38.560 | and it's, I think it will be challenging, like,

00:52:40.720 | to do both, like, and maybe, maybe you can.

00:52:43.040 | I don't know, like, we do see these numbers right now,

00:52:45.680 | raising above $100 million, even without having a product.

00:52:49.320 | You can see these.

00:52:50.520 | - It's excellent advice.

00:52:51.400 | And I think what's been amazing,

00:52:53.320 | but also kind of challenging is you, you know,

00:52:55.920 | we're like trying to forecast, you know,

00:52:57.840 | okay, well, what, where are these things going?

00:52:59.440 | 'Cause I mean, in the initial weeks,

00:53:00.880 | I think us and all the, you know,

00:53:02.440 | investors in the company that we're kind of

00:53:03.640 | sharing this with, we were, it was like,

00:53:05.040 | this is cool, okay, we added 500K.

00:53:07.320 | Wow, that's crazy.

00:53:08.160 | Wow, we're at a million now, okay.

00:53:09.000 | And you kind of, you know, most things,

00:53:11.160 | you have this kind of the tech crunch launch of initiation,

00:53:13.640 | and then the thing of sorrow.

00:53:15.280 | And this is, this is just, you know,

00:53:17.080 | we haven't, if there's going to be a downturn,

00:53:19.200 | it just, it's just not coming yet.

00:53:21.160 | Now that we're kind of looking ahead,

00:53:22.640 | we're six weeks in.

00:53:23.600 | So now we're getting enough confidence in our convictions

00:53:25.880 | to go, okay, this, this kind of seems to be,

00:53:28.280 | this seems to be the trend line.

00:53:29.720 | - I'll tell you what other reason why I think like,

00:53:32.000 | where is Jasper?

00:53:33.640 | - They actually just announced some new numbers recently.

00:53:35.960 | Like they, you know, they're still surviving.

00:53:37.880 | They have gone down a lot.

00:53:38.720 | I think that the peak that I heard was 100 billion ARR.

00:53:41.920 | - Right.

00:53:42.760 | - And then they died.

00:53:43.600 | - And now, you know, there's like tens of these.

00:53:45.520 | So I think like, and their success was phenomenal.

00:53:49.920 | Like what I see at Bolt.

00:53:51.840 | And I think like, if you want to keep that probably,

00:53:54.920 | who am I?

00:53:55.760 | Like, I'm just like giving my two cents.

00:53:57.080 | - No, I think you're right though.

00:53:57.920 | - You need to focus because you are like,

00:53:59.520 | you are going to see like weeks.

00:54:01.920 | I think that you're disrupting their market

00:54:03.800 | and you open sourced some of it

00:54:06.000 | and they have containers, I believe.

00:54:08.200 | - Yeah.

00:54:09.040 | - And you need to fight.

00:54:09.880 | Like, I can tell you that when we open,

00:54:11.240 | so like I share with you like a small competitor,

00:54:13.560 | but I can tell you, I have a friend

00:54:15.280 | who has built a billion dollar company and more.

00:54:19.400 | Like when we released Alpha Codium,

00:54:21.000 | he like sent me a private email asking,

00:54:23.400 | what the fuck did you just do?

00:54:25.160 | Like, why did you release that?

00:54:26.360 | - You should have kept it.

00:54:27.280 | - Yeah, like you released that open source.

00:54:28.960 | Like, you know, I'm thinking build some stuff

00:54:31.760 | and now I can do that much more easily.

00:54:34.400 | I can tell you my answer.

00:54:35.480 | And I think, I thought that maybe you will answer as well.

00:54:37.520 | Although like, I think Bolt already very promising.

00:54:40.280 | For us, like Alpha Codium one is like GPT one.

00:54:43.760 | I agree with you, like being open and open source,

00:54:46.440 | et cetera, really helps to improve the product community,

00:54:48.840 | et cetera.

00:54:49.680 | But at some point opening,

00:54:51.000 | I closed their GPT 3.5 or whatever.

00:54:54.000 | And that was part of my answer.

00:54:55.960 | Like Alpha Codium is the agent that is compatible

00:54:59.760 | with GPT one.

00:55:01.200 | And there is a lot to do for these agents

00:55:03.480 | to actually get that like moment that we had

00:55:07.000 | with GPT 3.5, et cetera, as agents.

00:55:09.720 | - Yeah, I think you did right.

00:55:10.600 | And I think it just comes back to,

00:55:12.600 | it kind of comes back to what GeoHot said, right?

00:55:14.360 | It's like, you know, if you want to win,

00:55:16.200 | there's no other option than out hustling everyone else.

00:55:18.520 | You know, and so I think that's kind of,

00:55:20.200 | and then out hustling in the sense,

00:55:21.560 | really meaning building the best product,

00:55:23.520 | building the best experiences.

00:55:24.840 | And so, and I think that's the only way

00:55:27.240 | kind of almost any route and, you know,

00:55:30.040 | kind of open source and some of the stuff

00:55:31.480 | just kind of burns the ships in a sense, right?

00:55:33.320 | And then maybe that's the simplest way of saying it,

00:55:34.840 | just you're burning the ships,

00:55:36.040 | but also it builds a lot of goodwill.

00:55:38.160 | Yeah, I mean, there's a lot, tons of benefits to it.

00:55:40.160 | - Salesforce are doing that, right?

00:55:41.440 | They're now going to be agent force or whatever.

00:55:44.120 | So you can also.

00:55:45.600 | - We're going to try to get Mark on the podcast

00:55:48.040 | and Decibel is good friends with Salesforce.

00:55:51.200 | Any parting thoughts, any, you know,

00:55:52.720 | trends that you're like super excited about?

00:55:54.720 | - If we're talking about trends,

00:55:56.160 | I go back to our original podcast

00:55:58.400 | where we talked about like the idea

00:56:00.720 | that the software world is like built

00:56:03.840 | from specs, tests, and code.

00:56:06.400 | And I think like you can see that like one dimension

00:56:10.400 | are company startups that are rethinking

00:56:13.440 | the entire like development environment.

00:56:16.840 | Like I think like Bolt, et cetera.

00:56:19.040 | And another dimension is like, where's their focus?

00:56:21.640 | Is it like on the spec, is on the test and on the code?

00:56:24.800 | And I think like, it's interesting to see that

00:56:26.640 | from that view.

00:56:27.960 | We'll see more startup and more amazing announcement

00:56:30.640 | like of new directions, new philosophy.

00:56:34.080 | So I think we'll see startup like focusing,

00:56:36.280 | let's build everything from the spec.

00:56:38.600 | To some extent, I would say that like Bolt is,

00:56:41.440 | from my understanding, you can say better,

00:56:43.080 | somewhere in the line between the spec and the code,

00:56:45.120 | because you start, like I saw your demos,

00:56:47.080 | you're trying to describe like things,

00:56:48.680 | like not just in one row,

00:56:50.000 | because you wanna like to look like you want in it.

00:56:52.080 | So it's on that edge between connecting

00:56:54.960 | between spec and code.

00:56:56.400 | And you see others like, I think like all the IDs,

00:56:59.080 | most of them are the new IDs or the fork are there.

00:57:02.760 | We are like more focused from the test

00:57:05.160 | and to the code and to the spec, et cetera.

00:57:07.840 | So these are trends, I think like we will see that.

00:57:10.680 | And I think like another dimension to consider is like,

00:57:14.360 | is it more for the highway AI, for the developers,

00:57:18.160 | maybe not even like the technical person

00:57:20.240 | or is it for the enterprise?

00:57:22.880 | And that also like gives you like different products.

00:57:26.200 | If they are aiming for different ICP,

00:57:28.840 | different ideal client profile,

00:57:30.880 | they will approach this triangle of spec and test and code.

00:57:34.440 | And that's like how I see the world.

00:57:36.200 | And what I'm noticing is that we're seeing more and more

00:57:39.600 | of those new startups, new interfaces

00:57:42.640 | that are not focused on code.

00:57:45.000 | For example, talking more about the spec,

00:57:48.400 | talking more about the testing.

00:57:50.160 | Eventually I think that that's where the world is going to.

00:57:53.360 | Like the code is going to be there

00:57:56.000 | and there will be developers, et cetera.

00:57:58.200 | But as agent improves and capabilities of the LLMs

00:58:03.000 | and integrations to different parts of the environment,

00:58:06.640 | development environment, we're going to see more and more

00:58:08.800 | like focusing on the spec and the test.

00:58:11.360 | Basically, like these two might unite,

00:58:13.440 | the spec and the test,

00:58:14.280 | because you can say that tests are runnable specs,

00:58:18.680 | right, to some extent.

00:58:19.680 | So like that's another way to look at it.

00:58:21.720 | Yeah.

00:58:22.560 | - Literally on the slide here,

00:58:23.680 | runnable tests right here.

00:58:25.120 | - Yeah, I'm consistent.

00:58:25.960 | - It's all consistent.

00:58:26.800 | - Yeah, look, I talked about system one and system two,

00:58:29.560 | like more than a year ago.

00:58:30.960 | - Yeah, yeah.

00:58:31.800 | - And now with O1, people are talking about system one.

00:58:36.200 | But I think we'll talk about it again

00:58:37.320 | because I think they're totally, totally wrong

00:58:39.720 | about O1 being a system two.

00:58:42.040 | So like, it is now like in the hype

00:58:44.760 | or whatever talking about that.

00:58:46.480 | But I think like the agents are the one

00:58:48.680 | that will take us towards system two.

00:58:51.600 | And the more they are aware of their environment

00:58:55.440 | and aware of that sometimes they don't know

00:58:58.080 | what they don't know,

00:58:59.360 | then we'll really get to system two.

00:59:01.320 | But that's like a deeper discussion.

00:59:02.800 | - It's a deeper discussion.

00:59:04.200 | Love the philosophy talk that we had last time as well.

00:59:06.680 | All right, so we're back onto Bolt

00:59:08.600 | and it's more hard to leave for another interview,

00:59:11.680 | but we were just talking about like the,

00:59:12.960 | what happened post-launch, right?

00:59:14.480 | And I held this sort of emergency council of advisors

00:59:19.480 | for you because we had never seen this before.

00:59:22.520 | And I was like, okay, I'm going to call

00:59:23.480 | all the smartest people I know to join this thing.

00:59:25.960 | - Which was extremely helpful.

00:59:27.040 | I'm so appreciative.

00:59:28.320 | There's been a handful of me.

00:59:29.760 | - You made a couple, you made one hire out of that.

00:59:32.840 | - Yeah, 'cause it was like,

00:59:34.120 | I think I can't remember where we were at kind of ARR wise

00:59:37.320 | when I had messaged you.

00:59:38.640 | - It was like, you messaged me at like two or three.

00:59:41.440 | - Okay.

00:59:42.280 | - And then by the time we got everything together,

00:59:43.640 | it was four.

00:59:44.840 | And then, yeah, now it's at.

00:59:46.800 | - Since Eric sat down five minutes.

00:59:48.800 | (both laughing)

00:59:50.400 | - But I mean, it sounds like you've accelerated

00:59:52.600 | 'cause you told me it was like 100K, 200K a day.

00:59:55.600 | And now it's accelerated.

00:59:56.840 | - Yeah, this past, I mean,

00:59:59.440 | every week has been kind of a blowout week as far as.

01:00:03.000 | - Is it TikTok?

01:00:03.840 | - We're digging into the degree that we can.

01:00:06.560 | Kind of just like where all this stuff's coming from.

01:00:08.480 | I mean, there's a ton of word of mouth, right?

01:00:10.480 | So that you can't,

01:00:11.320 | which you can't just like look by refer, right?

01:00:13.400 | So there's a ton of direct.

01:00:14.480 | But yeah, I mean, there's a lot of TikTok.

01:00:16.720 | There's a ton of YouTube.

01:00:18.360 | It's kind of, I think, been a sensation

01:00:20.360 | in the sort of like, you know, entrepreneurial,

01:00:22.920 | build your own SaaS, indie hacker,

01:00:24.920 | you know, even developer circles.

01:00:26.920 | And I think too, like our team's been doing

01:00:28.360 | a really good job.

01:00:29.320 | Our folks just kind of like flipped a switch.

01:00:30.960 | And like, we, you know,

01:00:31.960 | people were just working through the weekends

01:00:33.520 | or whatever to get stuff fixed.

01:00:34.920 | And so the product,

01:00:35.760 | and you'll see people say this online.

01:00:38.120 | Like today, there was a tweet.

01:00:39.040 | Someone was like, yeah, I tried this like the first week

01:00:40.920 | and like couldn't get whatever to work.

01:00:42.240 | Came back today, you know, six weeks later,

01:00:45.640 | and this is ridiculous.

01:00:46.960 | Like, this is so good, right?

01:00:48.320 | And so I think there's been an incredible amount

01:00:50.080 | of improvement to the product, to the agent,

01:00:52.360 | also to like the underlying models too, like Sonnet,

01:00:55.240 | you know, that just happened to do an update on,

01:00:57.360 | you know, with their release like a couple of weeks ago.

01:00:59.880 | And so, you know,

01:01:00.720 | when we put our new agent online and the new Sonnet,

01:01:03.080 | we saw a huge bump in conversion just based on that.

01:01:06.440 | And so, yeah, like, you know, we've gone at that.

01:01:08.560 | When we were chatting, that must have been three weeks ago,

01:01:11.200 | you know, maybe an average of 100K ARR per day.

01:01:13.960 | And, you know, this week I will see,

01:01:16.360 | so I've said this every week, but we'll see if it holds.

01:01:17.920 | You know, it's like kind of, we've,

01:01:19.120 | the past couple of days have been like, you know,

01:01:20.520 | half a million of ARR per day.

01:01:22.240 | Just, I mean, which, you know, insane.

01:01:23.720 | I mean, you know, I think today we had our peak,

01:01:26.640 | we've had peak traffic, you know,

01:01:27.800 | just kind of set the previous, you know,

01:01:29.280 | and that's kind of been, you know, every day this week.

01:01:30.880 | But anyways, yeah, I think things just continue

01:01:32.800 | to accelerate, which is kind of blowing my mind, you know,

01:01:35.080 | 'cause it's just the sheer numbers of this stuff

01:01:37.200 | are just like mind boggling.

01:01:38.840 | - I think you almost suffered from like the Twitter demo

01:01:43.160 | issues that other people had.

01:01:44.360 | Like the first time I saw Bolt,

01:01:45.880 | I saw the demo and I was like, oh, that's cool.

01:01:47.760 | Like I didn't go to try it because I was like,

01:01:49.400 | I've seen so many of these that it's like,

01:01:51.320 | I don't know if it's actually gonna work.

01:01:52.920 | And then two days ago, I signed up to use it.

01:01:55.800 | I was building a Luma replacement.

01:01:58.080 | I'm done with Luma.

01:01:59.520 | And I was like, man, this thing really works.

01:02:02.160 | I was like, and I already knew you, of course.

01:02:04.600 | And I was like, man, this thing really works.

01:02:06.240 | What the fuck?

01:02:07.080 | I was like, it's actually,

01:02:08.840 | I don't know if it's like the model,

01:02:10.080 | if it's like how you prompt it,

01:02:11.520 | but it's so good at coming up with like

01:02:13.120 | the simplest thing to implement.

01:02:15.320 | So the Luma example, right?

01:02:16.840 | So first I was like, create a RSVP page for an event

01:02:19.760 | and it created a wedding RSVP.

01:02:21.240 | I don't know if it's your fault.

01:02:23.040 | I don't know if you Bolted it of the model.

01:02:25.680 | And then I was like, well, now I need to have a way

01:02:27.520 | to create more events and added that.

01:02:29.400 | And then I was like, now it needs a way

01:02:30.480 | to like have an admin page to modify event.

01:02:33.080 | And maybe what I would have done as a developer,

01:02:35.080 | it's like, well, I'll create a different like admin view,

01:02:37.520 | you know, with all the events.

01:02:38.480 | And then I'll have like the front end thing.

01:02:40.080 | And instead what it did is like,

01:02:41.200 | it created like a admin view toggle on top.

01:02:43.800 | And then like just a pencil button on every page

01:02:46.280 | to edit them in line, you know?

01:02:48.080 | And that was it.

01:02:48.920 | And I was like, yeah, that works just as well.

01:02:51.160 | And like for the model,

01:02:52.520 | that's probably the simplest way to do it

01:02:54.120 | because it like limits the amount of files that are there.

01:02:56.800 | Can you talk just more about how much of this is like

01:02:59.100 | the model coming out with it,

01:03:00.120 | how much you're prompting it to kind of like be

01:03:02.440 | very like compressed and concise?

01:03:05.120 | - A ton of it is the model,

01:03:06.320 | but I think what's interesting though is,

01:03:08.520 | you know, kind of the baseline model,

01:03:09.800 | if I just like, if it's kind of like,

01:03:12.080 | try and put it into like a, you know,

01:03:13.520 | way if you had to quantify, you know, the effect is obvious.

01:03:17.320 | The model is like this sort of like 10X multiplier.

01:03:19.800 | How good the bottom line model is, huge, huge swing.

01:03:22.800 | And then kind of what you can do on top of that,

01:03:24.800 | you can squeeze out three, four X kind of more.

01:03:28.240 | And so that's kind of where the realm of, you know,

01:03:29.740 | prompt engineering and multi-agent approaches, et cetera,

01:03:32.420 | kind of kick in.

01:03:33.380 | And so I think with us, you know, our folks,

01:03:35.860 | like the guy on our side that, you know,

01:03:38.540 | led the web engineering,

01:03:40.020 | like that kind of our core technology for the past,

01:03:42.220 | you know, seven years here, you know,

01:03:43.660 | his name is Dominic Elm, based out of Germany,

01:03:45.820 | and he was one of the founding engineers of the company.

01:03:48.300 | Previous to StackBlitz,

01:03:49.140 | he actually was doing machine learning.

01:03:51.420 | And he basically had built a StackBlitz,

01:03:54.220 | like online IDE for machine learning.

01:03:56.320 | So think like a kind of like Google Colab sort of thing,

01:03:58.500 | or like Hugging Face has their kind of version of this.

01:04:00.700 | Back in 2016, it wasn't as much of a market for this stuff.

01:04:04.780 | But he had been doing a lot of, you know,

01:04:06.900 | training, you know, ML models and that sort of thing.

01:04:08.960 | So I guess, you know, as we began, you know,

01:04:11.500 | kind of digging into AI stuff over the past year,

01:04:13.660 | he's been kind of leading that off.

01:04:15.500 | And so a lot of it, I really attribute

01:04:17.780 | to Dom's specific angle,

01:04:20.540 | 'cause he has deep understanding of our technology

01:04:23.020 | and how it works 'cause he's, you know,

01:04:24.180 | led the engineering on WebContainer.

01:04:26.020 | But has a deep understanding of how these models work.

01:04:29.260 | Going and actually kind of writing out these, you know,

01:04:32.140 | whether it's like the prompt engineering aspect of it,

01:04:34.580 | or multi-agent or whatever have you, you know,

01:04:36.780 | that's sort of like that much context.

01:04:39.140 | And the other folks on the team are, you know,

01:04:41.260 | in the same sort of spot.

01:04:42.960 | They've been working on this stuff.

01:04:44.540 | I think we've been able to squeeze out a lot more

01:04:46.500 | than I've seen almost anything else out there,

01:04:48.660 | at least in the term of building web apps, at least.

01:04:50.920 | But I think it's kind of just because

01:04:53.900 | we have more context on a fewer number of heads

01:04:57.060 | at the company.

01:04:57.900 | So we can kind of connect the dots of it faster, you know?

01:05:00.100 | - Yeah, that's part of the issue

01:05:01.140 | with the whole raise a billion dollars thing.

01:05:02.680 | Like you actually run very lean

01:05:04.980 | and that's actually been to your advantage.

01:05:06.900 | - Totally, and I think we have to staff up

01:05:09.660 | 'cause we went from, you know, call it zero customers

01:05:12.940 | to, you know, 20, 30,000 kind of, you know, in six weeks.

01:05:17.060 | We have to have certainly more customer support,

01:05:20.020 | customer success stuff, et cetera.

01:05:21.700 | But you know, also just on non-engineering,

01:05:23.740 | we have to ramp up.

01:05:24.580 | But I do think that there's a,

01:05:26.740 | we saw this in the 2021 cycle, right?

01:05:28.700 | Where, you know, adding tons more people

01:05:31.540 | can be a thing that really hurts, you know, the company,

01:05:34.540 | because you can, it's just harder.

01:05:36.980 | It's really hard to manage lots of people.

01:05:38.540 | And that's, if you're a big enough company

01:05:40.340 | to warrant a certain headcount, 100%,

01:05:42.180 | you kind of have to do it, right?

01:05:43.660 | But I think for us, it's worked just to really,

01:05:46.660 | you know, grow the team slowly and intentionally.

01:05:49.420 | And so I think we're gonna take the same approach here

01:05:51.060 | at a bit of a faster clip than we were previously.

01:05:53.220 | But to me, that would just be general advice to startups

01:05:56.100 | is like slowly, intentionally,

01:05:57.860 | as fast as you can to meet demand or whatever.

01:06:00.980 | - Part of what I felt,

01:06:03.260 | like you're in a unique position to talk about,

01:06:05.220 | but also kind of what we went through in our call was,

01:06:08.940 | I have PMF, now what?

01:06:10.260 | Is kind of what I've been saying.

01:06:13.060 | And so like, I think the first answer

01:06:15.340 | is hire a data scientist.

01:06:16.780 | Because we have to sort of figure out like from our data,

01:06:19.900 | you're now sitting on a ton of different customers

01:06:22.340 | and we don't really know the different customer segments.

01:06:25.460 | You're starting to get an idea of churn.

01:06:27.060 | You're starting to get an idea of like segmentation.

01:06:29.020 | You already had data enrichment.

01:06:30.700 | One of my most interesting quotes from you

01:06:33.140 | from that session was that

01:06:34.460 | because you were selling to enterprise for so long,

01:06:36.700 | you had already set up all that stuff

01:06:38.860 | and it just like wasn't useful

01:06:40.700 | for a more sort of developer, bottom-up centric approach.

01:06:43.820 | - Yeah, and particularly because

01:06:45.420 | for the first time in the company's history,

01:06:47.140 | we're selling primarily to almost non-developers.

01:06:49.700 | And so everything that we've ever,

01:06:51.340 | all the playbooks we had, not relevant here, basically.

01:06:55.020 | Right?

01:06:55.860 | So one of our investors I talked with earlier this week,

01:06:59.420 | basically brought up a really great point,

01:07:00.660 | which is like, you are now a B2C company

01:07:03.780 | and how you operate needs to reflect that.

01:07:06.860 | - Which is what?

01:07:07.900 | I don't know.

01:07:08.740 | - Which is basically from an analytics perspective,

01:07:11.660 | like you're tracking everything, right?

01:07:13.860 | And to your point, you have people kind of around the clock

01:07:17.660 | slicing and dicing data to understand

01:07:20.220 | who are these people coming in,

01:07:21.820 | who are the types of people you actually want to retain

01:07:23.700 | versus people that you know are just gonna churn out.

01:07:25.620 | And that's okay 'cause they're not the actual like ICP

01:07:28.780 | that you're going for, right?

01:07:30.340 | When you're building stuff for enterprise software,

01:07:32.140 | the bar is a lot lower.

01:07:33.380 | And so kind of from the conversation before,

01:07:36.100 | one of the biggest,

01:07:36.940 | and this is kind of what we found with StackBlitz,

01:07:38.060 | which is kind of interesting.

01:07:39.340 | You mentioned it's like, as a startup,

01:07:41.340 | it's very hard to sell on prem, extremely true.

01:07:44.700 | But if you can do it, it's like the promised land

01:07:46.300 | because these companies, they're Fortune 500s,

01:07:49.340 | they can write really large checks.

01:07:51.180 | And so when you're going and selling to them,

01:07:52.700 | it doesn't matter so much like on your website.

01:07:54.300 | Sure, you want to track the conversion

01:07:55.940 | to the enterprise contact form or whatever, right?

01:07:58.580 | But what actually really matters

01:08:00.340 | is like a lot of human touch points of,

01:08:04.540 | hey, we want to have a quarterly call after,

01:08:06.340 | just getting installed and stuff.

01:08:07.620 | There's a whole playbook for that.

01:08:08.660 | And you need to hire sales engineers

01:08:09.780 | that can be on the ground floor

01:08:10.980 | and helping people install it.

01:08:12.100 | Then after that, you got to, okay,

01:08:13.740 | how do we make sure they're kind of constantly successful?

01:08:15.660 | Because you can't access,

01:08:17.820 | like we can, our enterprise customer instances,

01:08:20.460 | we have no idea how often they're using them.

01:08:22.220 | Why?

01:08:23.040 | Because the whole point is that

01:08:23.880 | we can't see what they're up to for a good reason, right?

01:08:27.740 | Like they need to own their data.

01:08:29.700 | And so the way,

01:08:30.540 | it's actually much a very complicated problem

01:08:32.540 | of how do you have like build relationships

01:08:34.540 | where everyone's getting on calls,

01:08:35.780 | they can share kind of the telemetry

01:08:36.980 | that they can see within their instance,

01:08:38.860 | and you can kind of extrapolate that,

01:08:40.380 | make sure they're happy and successful.

01:08:42.020 | So there's a whole art of that,

01:08:44.380 | of doing enterprise well,

01:08:46.140 | that we've gone and done and closed these folks,

01:08:49.700 | totally unrelated to doing B2C,

01:08:51.380 | completely unrelated for the most part.

01:08:53.620 | So anyway, so that, as a company,

01:08:55.940 | we're kind of reorienting, you know,

01:08:57.740 | our focus on, okay,

01:08:59.180 | going and actually really leaning in on analytics,

01:09:01.620 | whatever have you.

01:09:02.460 | And fortunately, like my co-founder and I,

01:09:04.100 | the enterprise business at StatWoods

01:09:05.860 | was the first time we had ever done enterprise.

01:09:08.460 | Primarily, like thanks to the company we did before,

01:09:10.140 | was B2C.

01:09:10.980 | Like we were selling people courses

01:09:12.340 | on how to do web development, basically, right?

01:09:14.540 | So a lot of the skillset that, you know,

01:09:17.340 | I had built up there,

01:09:18.420 | I'm able to pull that back off the shelf,

01:09:20.780 | dust it off, sharpen the blade,

01:09:22.100 | and you know, we're doing email marketing,

01:09:24.700 | we're doing live streams, you know?

01:09:26.500 | So anyway, so that's, it's kind of cool to, you know,

01:09:29.180 | be shifting back to some of the,

01:09:30.940 | where we cut our teeth on back in the day.

01:09:32.500 | - How did you pick the pricing?

01:09:33.980 | - Because I had to pay.

01:09:35.180 | (laughing)

01:09:37.220 | - That's fantastic, that's fantastic.

01:09:38.860 | - Yeah, you want to like, slightly like,

01:09:40.620 | "Ugh, ugh."

01:09:41.460 | - It's like, you're running out of tokens, dude.

01:09:44.900 | I was like, "Fuck, I'm running out of tokens?"

01:09:46.780 | It's like, "I don't want to run out of tokens."

01:09:48.460 | But there's like five different tiers, right?

01:09:50.780 | Which are kind of like token-based and capacity-based.

01:09:53.700 | How do you reconcile that and the consumer side,

01:09:56.380 | where maybe the consumer doesn't even really need to know

01:09:58.860 | what a token is, right?

01:10:00.140 | Like on that, like your mom probably doesn't really care

01:10:02.620 | what an AI token is.

01:10:03.740 | How did you structure it to start?

01:10:05.220 | How did you come up with that?

01:10:06.140 | And then maybe ideas that you have to like improve

01:10:08.980 | or like modify that.

01:10:10.020 | - Totally, yeah.

01:10:10.860 | So when we first launched with StackBlitz is like,

01:10:13.140 | we were an enterprise play, right?

01:10:14.860 | And so when we launched in 2017,

01:10:17.220 | I think we tried pricing 2018 or 2019,

01:10:19.060 | like it was free for a long time.

01:10:20.180 | And then we had a $9 plan and that was just the way it was.

01:10:23.020 | It was kind of like our $1.50 hotdog at Costco.

01:10:25.820 | It's kind of like this, you know, just low price,

01:10:28.580 | just, you know, it wasn't the primary rev driver.

01:10:31.020 | And we just wanted to, you know, say,

01:10:32.140 | "Hey, pay for some more storage

01:10:33.460 | and private projects or whatever."

01:10:35.060 | And so we went to launch bolt again,

01:10:36.180 | like our expectation was,

01:10:37.620 | "Hey, we'll probably get a good number of people

01:10:39.460 | that'll sign up and be excited about it.

01:10:41.140 | And, you know, we're not too concerned."

01:10:42.780 | You know, we're just, we're just not,

01:10:44.060 | we were unprepared for the tsunami that hit.

01:10:46.540 | And so after going online the first week,

01:10:48.500 | we were like, "Wow, this is cool."

01:10:49.500 | There's, I mean, it just kept growing.

01:10:51.420 | And then once we hit week two, I mean, we were just,

01:10:54.100 | nine bucks was, I mean,

01:10:55.140 | it's like the cheapest AI coding thing you can get,

01:10:57.820 | maybe other than copilot,

01:10:58.700 | but like we were overrun by support tickets

01:11:02.180 | and just the sheer volume of people coming in.

01:11:04.620 | It's just laws of supply and demand.

01:11:06.020 | We were like, "Okay, this isn't,

01:11:07.900 | there's no way we can scale to meet this."

01:11:09.420 | Also, the people coming in

01:11:11.700 | are burning through their tokens

01:11:13.340 | and there's no way to actually like buy more of these things.

01:11:16.660 | And nine bucks is just,

01:11:17.580 | you can't get that much inference out of that.

01:11:20.140 | And so here's the other thing that's interesting about bolt

01:11:22.220 | compared to like something like copilot or whatever.

01:11:24.580 | And this kind of tied this,

01:11:26.020 | sorry, a little bit of a roundabout way

01:11:27.820 | to answer your question.

01:11:28.660 | But basically what we ended up at that moment,

01:11:31.220 | we ended up realizing is that when you use copilot,

01:11:34.180 | what it's sending up,

01:11:35.020 | it doesn't provide a lot of context of your code base.

01:11:37.260 | They try and reduce the amount of context

01:11:39.740 | as much as they can.

01:11:41.180 | And I think the origins of this stuff is they,

01:11:44.140 | everyone kind of wants this like low price point

01:11:45.820 | where it's like all you can eat.

01:11:46.780 | So it just kind of, it kind of feels like,

01:11:48.420 | 'cause it's like, it almost like Netflix.

01:11:50.460 | It's like, I'll pay a thing

01:11:51.380 | and then I can just do as much of the movie watching

01:11:54.500 | as I want.

01:11:55.340 | And I think that kind of mentality

01:11:57.860 | when these first AI products came, it kind of makes sense.

01:11:59.740 | They're like, "Okay, well, we don't want to meter it

01:12:01.300 | 'cause that doesn't feel good."

01:12:02.860 | But the problem is that then they're incentivized

01:12:05.500 | to not have it be able to keep,

01:12:07.740 | the more context you give it, the more it can do.

01:12:09.620 | And that's the magic of what we're doing with Bolt

01:12:11.860 | is we're giving it all the context we possibly can.

01:12:15.060 | And that's why you can go to it and say,

01:12:17.340 | "Make me an RSVP site."

01:12:19.100 | And it does it because it has context

01:12:21.060 | in the entire state of the application, et cetera, et cetera.

01:12:24.380 | And that's what makes it so accurate

01:12:25.780 | versus if you go to copilot and say that it'll be,

01:12:28.700 | it might punch out a React component

01:12:30.580 | that's the button to create the thing,

01:12:32.460 | but not actually more than that.

01:12:34.060 | So anyway, so, you know, and at the time

01:12:36.460 | when people had bought the $9 plan,

01:12:38.420 | they were like, "I want to give you more money.

01:12:40.620 | I want you to buy more tokens.

01:12:41.820 | How do I do that?"

01:12:42.660 | And so our team scrambled that weekend.

01:12:44.300 | We just turned it around and just, you know, we said,

01:12:46.380 | "Okay, well, what do we think is reasonable?"

01:12:47.940 | And we said, "Okay, so let's go."

01:12:49.780 | You immediately double the prices of the base tier

01:12:52.140 | because it's just not enough

01:12:53.740 | what people are getting on for nine bucks.

01:12:55.860 | So that'll be, that seems reasonable.

01:12:57.100 | It's kind of in line with everyone else.

01:12:58.420 | And then we added 50, 100 and $200 plans

01:13:00.660 | 'cause we're like, "That should be enough."

01:13:02.020 | And so, yeah, so that's kind of the origins of it.

01:13:04.540 | And it was people that use it, fall in love with it,

01:13:08.300 | and they want to use more of it.

01:13:09.780 | And the problem is the inference is expensive.

01:13:11.340 | And so we're not actually taking, you know,

01:13:12.780 | to date on the revenue we've done,

01:13:14.180 | we have not really taken a margin at all on this stuff

01:13:16.340 | 'cause we're just trying to put all the value back

01:13:18.020 | into the folks that are using the tool

01:13:20.020 | and just getting the D maximum amount of value out of it.

01:13:22.100 | But it's really key to kind of the magic of the experience.

01:13:24.580 | And so the other thing kind of worth mentioning

01:13:26.140 | is there's kind of the ARR number,

01:13:27.460 | but then you can also buy additional tokens, you know,

01:13:30.460 | just with usage-based billing effectively.

01:13:32.740 | And that's accounting for an additional 20, 30%

01:13:35.300 | of revenue that's coming to the company.

01:13:37.460 | People are actually using this to do their jobs.

01:13:40.580 | Like, think about a web development agency

01:13:43.220 | before this thing.

01:13:44.420 | They're going in using Figma to make a design.

01:13:46.580 | They have to pay the designer.

01:13:47.620 | They have to like punch that out into code, kind of,

01:13:49.260 | maybe like Copilot can help a little bit

01:13:50.700 | with punching out this thing, that.

01:13:51.940 | They're coming to this thing.

01:13:53.100 | There's just wild stories online where it's like,

01:13:56.100 | guy, a local bakery is like, "We need a website."

01:13:58.700 | He's like, "Okay, I'm going to charge you a thousand bucks."

01:14:00.980 | They're like, "Okay, that sounds great.

01:14:02.620 | Reasonable price."

01:14:03.700 | 30 minutes later, he's like,

01:14:04.820 | "Here's a deploy preview of your thing.

01:14:06.660 | How does that look?"

01:14:07.500 | They're like, "Wow, holy crap."

01:14:09.260 | - I'm not giving you a thousand bucks.

01:14:10.820 | (all laughing)

01:14:12.420 | - But they did.

01:14:13.260 | They were mind blown because they were like,

01:14:15.060 | "This usually takes months."

01:14:16.380 | - Yeah, yeah, yeah. - You know?

01:14:17.380 | So some of the biggest power users are people

01:14:19.300 | that build websites for a living

01:14:20.780 | because this is the alpha on this is insane.

01:14:23.940 | - That's almost like the gap, right?

01:14:25.420 | It's like, it used to be that if I ask you,

01:14:28.460 | before this, to do a website,

01:14:30.180 | and in 30 minutes you return to me

01:14:31.780 | and you give me something, I'm like,

01:14:33.940 | you know, you're probably just copying something else

01:14:35.660 | you've done before, you know?

01:14:37.140 | Versus now, it's almost like,

01:14:38.180 | it doesn't really matter how much time it takes you

01:14:39.980 | because everybody's going to be so fast with these things.

01:14:42.220 | It's more like the value.

01:14:43.500 | And that's why in your pricing, Tiara,

01:14:44.740 | was almost like, there's only really going to be like,

01:14:47.420 | either $20 a month users or like $1,000 a month users.

01:14:51.620 | You know, it's almost like,

01:14:52.700 | who's going to use the $50 a month?

01:14:54.940 | Because it's kind of like in between,

01:14:56.180 | between being infrequent user

01:14:58.020 | and being like a power user, you know?

01:14:59.980 | So yeah, it makes sense that you have like a big part

01:15:02.180 | of like on-demand on top of the tiers.

01:15:04.740 | - Yeah, and on the 50,

01:15:06.180 | there's actually a lot of people on the one.

01:15:07.500 | I think it's 'cause it's like enough to actually,

01:15:09.300 | like for developers that are using this

01:15:10.700 | to just kind of like punch out components

01:15:12.260 | or designs or whatever,

01:15:13.380 | kind of gets them enough for, you know,

01:15:14.860 | kind of in a given month or whatever.

01:15:16.620 | And so it's been interesting to just kind of see the,

01:15:18.900 | you know, the upgrades that happen.

01:15:21.140 | But what's been kind of cool about the product is it's,

01:15:23.420 | and again, I think this is kind of novel

01:15:25.380 | and this is, you know,

01:15:26.500 | us being maybe a little more transparent

01:15:28.100 | than we should be or something.

01:15:29.260 | But like, I suspect we're gonna,

01:15:30.660 | I think we're gonna see a lot more of this

01:15:32.340 | because we're hitting an inflection point.

01:15:34.420 | Coming back to the co-pilot thing,

01:15:36.220 | part of the problem before is that it didn't matter

01:15:38.380 | if you provided more context,

01:15:39.540 | the models just weren't good enough

01:15:40.500 | to know what to even do with it.

01:15:41.700 | That's not the case now.

01:15:42.900 | You know, just one, you know,

01:15:45.460 | story of like one of the first people,

01:15:47.340 | one of the first power users that adopted Bolt

01:15:49.420 | was this gal in Thailand

01:15:50.980 | who's a PM at a software banking company.

01:15:53.020 | And she had an idea for this app called ViralHooks.ai,

01:15:55.580 | which is basically, it's a tool that

01:15:57.660 | if you wanna make viral TikToks and stuff,

01:15:59.380 | it's like, what's the hook of the video

01:16:00.460 | to make people watch, right?

01:16:01.380 | And so basically she, you know,

01:16:03.260 | you can go and like see,

01:16:04.340 | it goes and extracts hooks from other people's videos

01:16:07.260 | and helps you with like, you know, AI to write your own.

01:16:10.460 | And she had originally, the week before Bolt launched,

01:16:12.940 | she put that on Upwork.

01:16:14.580 | And, you know, some, I think a developer in like Ukraine

01:16:17.740 | had quoted her, you know, $5,000.

01:16:19.420 | And it's gonna take like three months

01:16:20.500 | or something like that.

01:16:21.540 | Reasonable timeframe, right?

01:16:22.620 | For an app like that, reasonable price.

01:16:24.180 | The week after that, Bolt came out,

01:16:26.140 | she bought the $50 plan

01:16:27.820 | and she had the app built within a week or two.

01:16:29.940 | And so you're talking about like-

01:16:30.780 | - This is Bolt. - That's it.

01:16:31.980 | And it's beautiful.

01:16:33.300 | She did an incredible job, right?

01:16:34.940 | And so the numbers are wild.

01:16:37.940 | $5,000, three months to $50 in like a week.

01:16:42.940 | - Yeah, you gotta charge more, right?

01:16:45.220 | (all laughing)

01:16:47.260 | - But so it's kind of like, so, you know,

01:16:49.220 | there's people like when we've had a lot of people go,

01:16:51.500 | this pricing's insane.

01:16:52.580 | And we're like, well, we're not even taking really a margin

01:16:55.260 | at the moment on it, you know?

01:16:56.340 | But also, but when you compare that to the price

01:16:58.180 | of actually going and building,

01:16:59.900 | the cost of building quality software today,

01:17:01.620 | anyone who knows the price of building quality software,

01:17:03.980 | the alpha is obvious, right?

01:17:05.700 | It's a 99% cost production and 5X faster, you know,

01:17:09.900 | delivery time, you know?

01:17:10.780 | So anyway, so that's, I think we're one of the first products

01:17:14.140 | that have actually come out kind of proving that,

01:17:17.500 | you know, in a revenue way.

01:17:19.140 | To kind of underscore the point, as you can imagine,

01:17:21.580 | we've had, you know, kind of venture capital firms

01:17:23.740 | kind of reach out and kind of, you know,

01:17:24.940 | curious to kind of, you know, what we're up to or whatever.

01:17:27.180 | And so, you know, one of the most, you know,

01:17:28.860 | there's kind of, one of the most notable ones

01:17:32.300 | or whatever reached out, so we kind of sent them,

01:17:34.180 | you know, kind of our numbers.

01:17:36.060 | Actually, it was the investor update, Sean,

01:17:37.820 | that I think, you know, the one you saw,

01:17:40.340 | kind of gave them a snapshot of it.

01:17:41.500 | And one of their analysts accidentally replied all

01:17:44.300 | on what we had sent them with the analysis.

01:17:46.860 | And so on this part there, one of the things they said was,

01:17:50.060 | we haven't seen anything that's kind of eye-opening

01:17:52.660 | to see people going to $200 tier on this sort of thing.

01:17:56.100 | Haven't seen anything else like that in the space.

01:17:58.020 | 'Cause I think this is very new

01:17:59.420 | because of the new model capabilities, right?

01:18:01.260 | Where people, you know, it makes sense.

01:18:02.940 | Like you're willing to pay more money for this stuff, so.

01:18:05.100 | - This is something I've talked about before

01:18:06.860 | in terms of matching the dollar amount of spend

01:18:10.340 | to the capabilities of the AIs.

01:18:12.300 | The chart that I published in the past was, you know,

01:18:14.900 | OpenAI has like five levels of AGI-ness

01:18:17.860 | and level one is sort of like a chat boss.

01:18:20.460 | Level two is reasoning.

01:18:22.020 | Level three is agents.

01:18:23.140 | Four is organizations.

01:18:24.540 | Five is something super, super human.

01:18:27.220 | I don't remember what the exact levels are,

01:18:28.900 | but you can sort of each match each of them with like tiers.

01:18:32.060 | Like $20 is like the ChatGPT tier.

01:18:34.380 | $200 is where you're at.

01:18:35.980 | $2,000 is higher, 20,000, 200,000, right?

01:18:39.580 | You can see levels where it makes sense.

01:18:42.380 | I think BrightWave is also there, by the way.

01:18:43.820 | Like, I don't know what BrightWave charges,

01:18:45.220 | but it's higher, right, than a ChatGPT.

01:18:47.940 | And like, you have to deliver more value for that,

01:18:51.180 | but you can do it now.

01:18:53.020 | So then why not?

01:18:53.860 | I mean, everyone should do it.

01:18:54.780 | (laughing)

01:18:56.140 | - I think we're going to see a lot more of this.

01:18:57.700 | I think we're going to see,

01:18:58.540 | and I think, you know, for AI, Cogen specifically,

01:19:01.740 | this is the first moment where I think that

01:19:04.220 | there's been that moment where it goes from zero to one,

01:19:07.300 | where it's like, yep, the price point, you know,

01:19:09.420 | the value, the value is so,

01:19:12.020 | like of what you can get out of these things

01:19:13.380 | is so much higher than it was, you know,

01:19:16.220 | three, six months ago that I think we're going to see,

01:19:19.700 | I think we're going to see a lot more of this.

01:19:20.940 | Like we might, you know, Bolt is, I think,

01:19:23.420 | one of the first things, but yeah, I mean, it's just,

01:19:27.460 | to me, it's inevitable that we're going to see

01:19:29.940 | many more things kind of leveraging this sort of use case

01:19:33.620 | and the amount of efficiency you can get

01:19:35.180 | out of using these systems, right?

01:19:36.700 | So, yeah.

01:19:37.940 | - Yeah, because, I mean, the Bolt arbitrage

01:19:39.620 | would be, quote, the price based on the query.

01:19:42.060 | - Yeah, you're selling high value tokens.

01:19:43.700 | - Yeah, it's like, hey, it's like your mom is like,

01:19:45.900 | you wouldn't charge your mom $2,000 to tell her stories,

01:19:48.420 | but like, you know, this person doing an app

01:19:50.740 | and like a product on it, you gotta pay more, you know?

01:19:54.380 | But it's hard right now.

01:19:55.580 | I understand it's like, it's really hard to figure out

01:19:58.100 | how much you can push it,

01:19:59.580 | how much value the person will get out of the thing.

01:20:02.980 | - So I want to riff a little bit on stuff like this, right?

01:20:05.500 | I think you nailed a lot with the design system.

01:20:07.420 | You know, one of the differences between open source Bolt

01:20:09.700 | and the one that you have is actually like,

01:20:11.900 | you spend a lot of time on the design system, I think, right?

01:20:15.820 | Most things just look great when they come out,

01:20:17.820 | but then there's also a whole backend portion

01:20:19.700 | that they need.

01:20:20.540 | Was that a challenge?

01:20:21.420 | Is there anything that you sort of like figuring out

01:20:23.180 | that you want to riff on?

01:20:24.260 | - Yeah, so I think one of the main things,

01:20:26.140 | I think you hit the nail on the head,

01:20:27.420 | which is, you know,

01:20:28.460 | and kind of going into putting Bolt online,

01:20:30.380 | we originally, again, we've been selling to developers.

01:20:32.340 | And so we were kind of like, this is a tool for prototyping

01:20:34.220 | and they'll download their code.

01:20:35.820 | But what we ended up finding in the early user testing

01:20:37.820 | was how important the deployment story was

01:20:39.500 | and how, and this is something you said to me specifically,

01:20:42.380 | you're like, backend, this needs to like,

01:20:44.660 | backend needs to be part of this,

01:20:45.780 | like logging in, like off.

01:20:47.620 | Just to triple confirm, you're dead right.

01:20:50.460 | That has been the absolute number one thing

01:20:54.220 | that folks coming to Bolt, you know, are looking to do

01:20:56.500 | is build a real app with a backend, with billing.

01:20:58.420 | And so one of this guy, Mauricio,

01:21:00.660 | he's one of our power users.

01:21:02.060 | He's like, there's three things that like every app

01:21:04.740 | that I'll ever want to build in Bolt

01:21:05.980 | and any of these other people in this community,

01:21:08.060 | there's three things, a database, auth, and payments.

01:21:11.940 | Those three things, right?

01:21:12.940 | And so that's-

01:21:14.500 | - Admin dashboard.

01:21:15.620 | - We can do that pretty decently, pretty decently.

01:21:18.220 | - As in every database needs a WP admin.

01:21:21.500 | - Yes, yes, correct.

01:21:22.380 | Totally, totally.

01:21:23.500 | And so, yeah, today I think like Viral Hooks, for example,

01:21:26.060 | I think she's using Firebase for auth and database

01:21:28.940 | and that sort of thing.

01:21:30.100 | You know, so I think Firebase and Superbase,

01:21:31.740 | those are the two things that just work incredibly well.

01:21:35.140 | And so that's actually the point where we're at now,

01:21:36.940 | where, you know, right now it's, you know,

01:21:38.860 | folks have to still, you know, kind of go to Superbase,

01:21:41.100 | manually spin up a thing, come back to Bolt,

01:21:43.100 | but the thing that, you know, it's like

01:21:44.500 | that sort of processing thing with Firebase.

01:21:46.820 | Each of those products are gonna have

01:21:47.900 | their own little quirks.

01:21:48.740 | You have to, there's like kind of steps, right?

01:21:50.580 | And so-

01:21:51.580 | - Boltbase.

01:21:52.420 | - Yeah, Boltbase, yeah.

01:21:55.420 | I think, yeah, I think initially we're like,

01:21:58.020 | okay, there should just be a way to like,

01:22:00.820 | for Bolt to just go and spin up these things

01:22:03.300 | on their behalf and just, you know,

01:22:05.620 | both of them have APIs to do so.

01:22:07.580 | - I'll go even further, like have like pre-warm instances

01:22:11.100 | that you just assign, like it's already spun up, right?

01:22:14.060 | So it's like a kind of serverless feeling,

01:22:16.420 | even though it's like not really, but like, yeah,

01:22:18.780 | just pre-warm and then just kind of assign it

01:22:20.460 | whenever someone like spins up an app.

01:22:21.780 | - That's a really great point.

01:22:22.620 | Yeah, just keep one Firebase in the hopper, basically.

01:22:25.540 | - One, 10, 100, I don't know.

01:22:27.540 | More generally, this is what I felt

01:22:29.780 | that I wanted to do on our call,

01:22:31.700 | which is like when you have PMF,

01:22:33.340 | yes, you want to invest some time in like

01:22:35.180 | understanding your customers and do a data analytics

01:22:37.580 | and like tighten things up in general,

01:22:39.220 | like tighten up the pricing,

01:22:40.620 | tighten up the cost and all that.

01:22:42.460 | But then like, you also have to work on like,

01:22:44.060 | what is next, like the next level in growth,

01:22:46.900 | like you can still inflect.

01:22:48.780 | I don't know what that is, but you know,

01:22:50.540 | I wanted to keep pushing you and I don't know if I did,

01:22:54.060 | mostly 'cause I was serving as facilitator on that call.

01:22:57.540 | That's what I think.

01:22:58.380 | Like, I think you got to still keep pushing the frontier

01:23:00.380 | and I don't know what it is, but like, you know,

01:23:02.580 | I want to hear what you got thinking about.

01:23:05.060 | - I think there's, you know,

01:23:06.580 | we've addressed just a lot of the low-hanging P0 stuff

01:23:09.300 | then and we've actually seen, we've kind of the,

01:23:11.940 | yeah, there's key moments where it's just kind of like

01:23:14.300 | been going like that, which has been cool.

01:23:16.420 | 'Cause it's like, okay, well, we're just getting started.

01:23:18.500 | This is just the fixing obvious things part.

01:23:21.620 | Fundamentally, I think what a lot of people

01:23:23.060 | are coming here to do is just,

01:23:24.340 | how can we just make it faster to go from idea to production?

01:23:27.300 | And a lot of it is like when I have to go

01:23:29.900 | to Firebase Superbase, spin something up,

01:23:31.740 | run a migrate, you know, like add a table,

01:23:33.420 | it's like the agent can do that, you know?

01:23:35.340 | So that stuff should be baked in.

01:23:36.780 | Yeah, and same thing with the deployment side.

01:23:38.580 | It's like, right now it's going to Netlify,

01:23:39.900 | but people have to create a Netlify account

01:23:41.180 | and go and do that, right?

01:23:43.020 | And so I think one of the things we're gonna end up doing

01:23:44.740 | here is just having the hosting be baked in.

01:23:46.740 | And so I've been talking with Matt over at Netlify

01:23:49.580 | about this, 'cause they actually have a way to kind of-

01:23:51.180 | - Yeah, the white label stuff.

01:23:52.020 | - Yep, the white label stuff.

01:23:52.940 | And so, 'cause people are just going to make a website,

01:23:55.380 | you know, and so it's-

01:23:56.940 | - I mean, that means also

01:23:57.940 | you take over domain registration.

01:23:59.860 | - Can you imagine, right?

01:24:01.340 | A couple of months from now, you come to this thing,

01:24:02.660 | you're like, "I want to make an RSVP site," right?

01:24:06.660 | And it's like, "Great, do you have a name for it?

01:24:08.780 | Do you want a domain?"

01:24:09.980 | You're like, "I don't know a name."

01:24:11.060 | It's like, "Well, here's like 10 options

01:24:12.220 | and the dot coms are available.

01:24:13.420 | Does any of them look good?

01:24:14.380 | Yep, that one does.

01:24:15.340 | Okay, do you want to buy it?

01:24:16.660 | Okay, great.

01:24:17.500 | It bought, the DNS is pointed at the thing.

01:24:20.140 | Should we start building this?

01:24:21.180 | Okay, does this look good?

01:24:22.420 | Yep, okay.

01:24:23.540 | Am I okay to push this to prod?

01:24:25.100 | Yep, that looks good."

01:24:26.300 | You know, like that's without leaving the product, right?

01:24:29.260 | - So to me, like Itamar was the first to actually say like,

01:24:31.580 | "You are the new Wix."

01:24:32.660 | I never, I personally never thought about it that way.

01:24:34.780 | Wix is a $10 billion company.

01:24:36.580 | Where you want to go, you know?

01:24:37.860 | 'Cause you still have a choice here.

01:24:39.300 | - From what we're hearing from the folks using the product,

01:24:41.660 | I think, I don't even think Wix is even able

01:24:44.780 | to solve their need, you know?

01:24:46.300 | - Yeah.

01:24:47.140 | - But not to say that we don't want to, you know,

01:24:48.700 | that that, what you're saying is now what we want,

01:24:50.340 | but I mean, yeah, like,

01:24:51.580 | I think we want to solve folks' problems.

01:24:53.220 | And I think that there's a huge gap in the market

01:24:55.540 | of being able to build, you know,

01:24:56.900 | kind of more sophisticated, high quality software,

01:25:00.420 | like websites in a way that,

01:25:02.580 | for someone who's a non-engineer.

01:25:04.340 | And so I think there's a huge market for that.

01:25:05.900 | And obviously, even if you're trying to build

01:25:07.020 | a wedding website, yeah, this is easier and faster, right?

01:25:10.780 | So I love it.

01:25:12.180 | You know, again, coming to the origins of why Albert,

01:25:14.780 | my co-founder and I are doing this is,

01:25:16.500 | we've always just loved building stuff on the web.

01:25:18.500 | It's like this, this is the tool,

01:25:20.180 | from what, even when Stack was just the IDE interface

01:25:22.860 | to the technology.

01:25:23.860 | It's like, this is the thing we wish we had

01:25:25.620 | when we were 13 years old, you know?

01:25:27.060 | And with Bolt, oh my God,

01:25:28.380 | if this is the thing I wish we had when we were 13 years old,

01:25:30.620 | I'm so glad that my daughter's gonna have this thing,

01:25:32.660 | you know?

01:25:33.500 | So anyways, yeah, I think it makes me pretty, pretty stoked

01:25:36.300 | that people are gonna be able to actually build

01:25:38.700 | amazing web applications

01:25:41.300 | that can do really sophisticated things, you know?

01:25:43.900 | So yes, I think the short answer is heck yeah.

01:25:45.860 | I mean, if that's sort of market,

01:25:47.700 | totally right up our alley.

01:25:49.180 | - One other angle that I wanted to pursue

01:25:51.020 | was also the other languages.

01:25:53.460 | You know, you're very JavaScript centric.

01:25:55.020 | We've talked about Python forever,

01:25:56.260 | Ruby maybe, is that important?

01:25:58.540 | You know, like the previous generation of site builders

01:26:01.620 | were mostly Ruby shops and then some PHP.

01:26:04.860 | Do we want to capture that?

01:26:06.380 | Or are we just like, you know, always been on JavaScript

01:26:08.380 | and just let JavaScript take over the world?

01:26:10.500 | - You know, I think we're certainly with great interest,

01:26:13.740 | interested in other languages.

01:26:14.980 | And we have like minimal support of Python

01:26:17.420 | and some C++ stuff in web container

01:26:19.700 | that you can like run or whatever.

01:26:21.220 | I think especially with the stuff we're seeing though,

01:26:23.180 | it's the language is kind of ancillary to the thing.

01:26:27.300 | - Well, there's the ecosystem of like,

01:26:29.140 | I want to end up with a code base

01:26:30.620 | that I can hire humans on

01:26:32.180 | to do the stuff that Bolt cannot do.

01:26:33.900 | - Yeah, true.

01:26:34.740 | And I think in that sense,

01:26:35.580 | like the JavaScript Node.js ecosystem

01:26:37.980 | is huge and well-established.

01:26:39.780 | So I think you'll certainly be able to get people

01:26:41.620 | to work on this stuff.

01:26:42.460 | And I think the only thing that would be missing

01:26:43.580 | is it's like, are you building web apps

01:26:45.460 | where a lot of the functionality

01:26:47.580 | is only in libraries that are in Python or something, right?

01:26:50.700 | And I think just kind of seeing the applications

01:26:52.860 | that are being built here,

01:26:54.540 | I think that'd be like data science

01:26:55.820 | and like ML and that sort of thing.

01:26:57.260 | And so that's, we're not seeing a lot of that stuff.

01:27:00.180 | And then, but I think that's like worth like

01:27:01.780 | kind of a more generic approach

01:27:02.900 | is like what Repl.it's doing,

01:27:03.860 | where they're spinning up real VMs.

01:27:05.700 | You can kind of run anything.

01:27:07.260 | And I think they started off with like doing Python service.

01:27:09.180 | Actually, I haven't tried their new agent stuff

01:27:12.220 | that's based on--

01:27:13.300 | - Repl.it agent, yeah.

01:27:14.540 | We're close friends.

01:27:15.820 | Repl.it has the database, the sort of live hosting,

01:27:20.300 | everything integrated that you're going to want to build.

01:27:23.460 | And I think you're on a collision course with them,

01:27:25.620 | to be honest.

01:27:26.660 | - We'll see.

01:27:27.500 | 'Cause I'm curious,

01:27:28.340 | and you're not the first person to say that.

01:27:31.300 | I'm curious to see how it shakes out.

01:27:33.060 | 'Cause I think the challenge is focus.

01:27:36.820 | You know, when you are,

01:27:38.980 | what's kind of the end goal that you're shooting?

01:27:40.620 | - Yeah, Repl.it's firmly for developers.

01:27:42.500 | You're positioning it for non-developers.

01:27:44.380 | Like that's--

01:27:45.380 | - Yeah, and even getting,

01:27:46.980 | even if focusing on a language or an ecosystem as well,

01:27:50.100 | 'cause again, the problem is that these things

01:27:52.020 | can just break in a million ways.

01:27:53.500 | And so part of the,

01:27:54.700 | a lot of the work in making the experience better,

01:27:57.300 | like how do you get,

01:27:58.140 | like how to make it,

01:27:58.980 | someone get an idea into their fingertips

01:28:00.660 | and live on prod, right?

01:28:02.660 | There's so much stuff in between there,

01:28:04.100 | and a lot of it is just errors that happen,

01:28:05.820 | and how do you handle those?

01:28:07.340 | And a lot of that comes down to

01:28:09.260 | having a giant database of common errors

01:28:12.820 | that you can maybe even fine tune stuff on

01:28:15.540 | at some point, right?

01:28:16.700 | So doing that on one ecosystem,

01:28:19.340 | you can move a lot faster

01:28:20.380 | than if you're trying to support

01:28:21.220 | a lot of different languages.

01:28:22.660 | However, it's to the point of,

01:28:24.420 | if you're kind of targeting developers,

01:28:25.780 | they may not need that level of kind of streamline,

01:28:28.180 | you know, thing.

01:28:29.380 | I think that's kind of where I see the main divergence

01:28:31.260 | is that we are unabashedly focused on this ecosystem

01:28:35.140 | for building web apps.

01:28:36.380 | - Got it.

01:28:37.220 | Yeah, you supported Vite forever.

01:28:38.620 | - Yeah, and so I'm very curious to see

01:28:40.140 | just how it all shakes out.

01:28:41.060 | 'Cause it's, I think what they're doing is actually,

01:28:42.980 | I mean, I'm very curious to see what Microsoft does.

01:28:45.580 | Because if anyone is good at giving out VMs,

01:28:48.700 | tying it to a coder, and putting AI in it,

01:28:51.500 | it sucks.

01:28:52.900 | He's got a cloud, he's got VS Code,

01:28:55.020 | they got code spaces, they're in open AI.

01:28:57.460 | Now they've got Anthropic in Copilot.

01:29:00.500 | I mean, I must imagine,

01:29:02.180 | I must imagine that they're cooking stuff over there.

01:29:04.580 | - We'll make sure to ask him.

01:29:05.700 | (laughing)

01:29:06.820 | Yeah, we have many friends from Microsoft

01:29:08.020 | listening to the pod.

01:29:08.940 | - So just to wrap, I don't know,

01:29:10.660 | is there anything else both related?

01:29:13.220 | I just have one personal question before we wrap the pod.

01:29:16.180 | - Maybe like just advice,

01:29:17.700 | now that you've been through this journey, right?

01:29:18.980 | Advice to your former self.

01:29:20.740 | - Okay.

01:29:21.580 | Yeah, at which point?

01:29:24.060 | - Yeah, advice to yourself.

01:29:25.940 | Thinking about, there are many founders out there

01:29:28.260 | with a business where they're like,

01:29:30.300 | they're working really hard at it, it's interesting,

01:29:32.580 | but it's not an AI business.

01:29:34.740 | And you kind of took the plunge to invest in this,

01:29:37.620 | and it worked out for you.

01:29:39.140 | Maybe a lot of people are like,

01:29:40.380 | okay, this guy got lucky.

01:29:42.220 | Obviously there's a little bit of luck in everything,

01:29:44.100 | but how do you improve your chances?

01:29:46.260 | Would you say go for it?

01:29:47.100 | Would you say everyone should go for it?

01:29:48.660 | How would you advise someone who was in your shoes

01:29:51.020 | and thinking about, maybe I should have a second product.

01:29:54.460 | Maybe I should take this experiment

01:29:55.900 | or maybe it doesn't work out.

01:29:56.780 | Like what's the calculus here?

01:29:59.300 | - Yeah, we were deeply skeptical going.

01:30:01.060 | I remember the conversation you and I had.

01:30:03.380 | You know, I was like this, I think there's something here.

01:30:05.820 | At that point we had built some amount,

01:30:07.020 | but I had waited a long time to give you the call.

01:30:09.500 | - I said, this is your moment.

01:30:10.580 | - Well, it was.

01:30:11.420 | - This is the fucking moment.

01:30:12.260 | - I remember specifically though,

01:30:13.100 | at the beginning of the conversation with Sean,

01:30:15.580 | he and I sat down at a coffee shop

01:30:16.980 | in SF and so I was kind of giving him the pitch of like,

01:30:20.740 | you know, I think we have,

01:30:21.700 | I think that I can't remember the exact framing I said,

01:30:24.460 | but it was obvious that Sean had heard

01:30:26.700 | a lot of people say this exact thing to him

01:30:28.300 | over the past year or two, which is like,

01:30:29.740 | hey man, we've got an AI play.

01:30:31.260 | Like this is our thing plus AI equals this.

01:30:33.860 | This could be crazy.

01:30:34.700 | And Sean, you gave me this like skeptical look.

01:30:36.540 | And I was like, I really think so.

01:30:38.620 | Okay, here's why, right?

01:30:39.900 | And I think that's, it's actually,

01:30:42.620 | I think that is internally being skeptical

01:30:46.380 | of just kind of going and jumping on hype trains is good.

01:30:50.340 | 'Cause it's like, I think you, you know,

01:30:51.980 | your focus and your time

01:30:53.180 | and what you're putting your weight into

01:30:54.700 | is the most important thing when you're a founder.

01:30:57.020 | I think for us, like we actually, again,

01:30:58.340 | like I had mentioned at the beginning of this, you know,

01:30:59.900 | we had tried Bolt and didn't see the results

01:31:02.180 | and that was like a two week sprint

01:31:03.300 | and we rolled it back, right?

01:31:04.460 | This isn't viable at this point.

01:31:07.100 | But then once we saw real tangible results

01:31:10.780 | of some of the new stuff, we're like, okay,

01:31:12.700 | that changes things.

01:31:14.100 | And I think a lot of it is, too,

01:31:16.060 | is going and finding that out for yourself

01:31:17.940 | and then going and talking to the smartest people you know

01:31:19.660 | with more domain knowledge on that stuff than you have

01:31:22.260 | and going, here's kind of what we found.

01:31:23.820 | Does this track?

01:31:25.060 | So when Sean and I met and he and I kind of,

01:31:28.300 | he saw it, we talked through it and he said,

01:31:29.940 | this is your moment.

01:31:31.100 | I specifically remember that.

01:31:32.340 | 'Cause I walked away from that and I was like,

01:31:34.380 | holy shit, this is it.

01:31:36.100 | Like this, you know, like Sean's at the intersection

01:31:39.100 | of web and AI and as like, you know,

01:31:41.740 | has one of the best perspectives on this stuff

01:31:44.020 | of anyone I know.

01:31:45.340 | That put a huge, you know, wind in our sails, honestly,

01:31:48.500 | of just like, okay, let's go and really,

01:31:51.340 | let's go and double down here because, you know,

01:31:53.860 | we had conviction before, but having someone

01:31:55.700 | who's in the space independently kind of verify

01:31:57.900 | meant a lot, you know?

01:31:58.860 | So.

01:31:59.700 | - It makes me uncomfortable, but thank you.

01:32:01.660 | - I mean it, I mean it, you know?

01:32:03.380 | And I waited, I waited until I was pretty darn sure

01:32:06.020 | it was not gonna be a waste of time to, you know?

01:32:09.660 | - Cool, well, that's all I have.

01:32:11.300 | - Yeah, and then on the personal side,

01:32:12.780 | you had a baby in April, you ran an Ironman in October.

01:32:16.220 | Now it's November.

01:32:18.020 | - He did Ironman while launching bolt.

01:32:19.980 | - Yeah, exactly. - I was trying to schedule

01:32:22.340 | the call for him and he goes like, nope, I'm running.

01:32:24.780 | - Sorry, I'm swimming.

01:32:25.620 | He's like, hey, I'm on the swimming session.

01:32:27.780 | - And for those who don't know, actually, I did not know,

01:32:29.620 | I don't even know the distance of an Ironman.

01:32:31.380 | - Yeah. - 13 hours.

01:32:32.820 | Your time was 12, 12, 12, 12, 12, 15.

01:32:35.420 | - Yeah, 12, 12, 12, 15. - Okay, oh.

01:32:37.420 | - You're like, give me my minutes, give me my minutes.

01:32:41.020 | - No, no, it completely depends on the course

01:32:45.020 | and just the person or whatever, right?

01:32:47.660 | But yeah, I mean, it's-

01:32:49.340 | - 2.4 KM open water swim. - That mile.

01:32:51.780 | Yeah, 2.4 miles. - Open water swim.

01:32:53.660 | 100 KM, 100 mile, 100 KM cycle.

01:32:56.700 | - I think it's 112 mile bike.

01:33:00.020 | - And then marathon.

01:33:00.860 | - Yeah, full 26.2 mile marathon, yeah.

01:33:03.300 | - Crazy. - It was-

01:33:04.140 | - Why?

01:33:05.300 | - Yeah.

01:33:06.140 | - And you were not like a super endurance athlete before,

01:33:08.900 | right, let's make this clear.

01:33:10.380 | - Yeah, kind of a wild thing.

01:33:12.260 | So back when we had our daughter in April,

01:33:16.180 | and at that time, the future of the company was,

01:33:19.260 | we were figuring out what are we going to do here?

01:33:21.500 | At that time, it was just prior to Bolt

01:33:24.300 | kind of getting kicked into the rebirth of it

01:33:26.700 | with the new models and stuff.

01:33:28.100 | And so I knew that it was going to be,

01:33:30.180 | having a child is,

01:33:32.780 | if you talk to anyone that's done that,

01:33:33.780 | you don't have a lot of sleep.

01:33:35.700 | There's a lot of, to be a great parent is a ton of work.

01:33:39.460 | And then also being a startup CEO

01:33:40.860 | where there's a lot of uncertainty or whatever.

01:33:43.260 | The way I've always found like when I have to go

01:33:44.780 | and kind of knock it out of the park

01:33:47.100 | in all aspects of my life is go and,

01:33:50.780 | yeah, just to make it all aspects of my life.

01:33:53.180 | And so I was, I just won, yeah, I woke up one day,

01:33:55.420 | I was like, all right, I'm going to do an Ironman this year.

01:33:57.540 | And I burned the ships, bought the,

01:34:00.420 | it costs a thousand bucks to do these, didn't know that.

01:34:03.740 | And just started, I never ran a marathon at that point.

01:34:06.780 | And so I think it was like 45 or 60 days after that,

01:34:09.740 | I ran a marathon.

01:34:10.700 | My brother-in-law, he's, that was even more insane.

01:34:13.140 | Two weeks before the marathon, I was like,

01:34:14.580 | hey, do you want to run a marathon in two weeks?

01:34:16.220 | He's like, sure.

01:34:17.220 | And just did it with me.

01:34:19.500 | He did not an endurance athlete either, right?

01:34:22.220 | But anyway, so yeah, so I was training,

01:34:24.940 | ended up getting a coach who's,

01:34:27.020 | you just go, you're kind of online, he's up in Marin.

01:34:29.420 | Great guy, was on the US Olympic team for triathlons.

01:34:32.660 | And when I told him, okay,

01:34:34.020 | I'm doing Ironman California in three months,

01:34:36.820 | he was like,

01:34:37.660 | are you insane?

01:34:41.340 | You know, like, what are you, you know,

01:34:42.620 | you'd ask for my opinion, but like, I just want you to know,

01:34:44.860 | I don't think this is a good idea.

01:34:46.460 | I think, you know, like you shouldn't do this, et cetera.

01:34:49.580 | And I ended up doing it, you know,

01:34:50.780 | I ended up getting it done.

01:34:51.780 | And so he was like, okay, like that's pretty bad.

01:34:53.540 | - But what makes you ignore expert advice here?

01:34:56.020 | Like most sane people would be like, okay,

01:34:59.460 | I mean, you know what you're doing,

01:35:00.300 | like I'll maybe wait a year.

01:35:01.980 | - So I think, and this is kind of the,

01:35:03.820 | and being a founder, right, it's all about like,

01:35:05.860 | if you, like I mentioned earlier, it's like,

01:35:07.540 | when we talk to people that worked on browser engines,

01:35:09.380 | they're like, you can't build what you're talking about.

01:35:11.780 | I think the job of a founder is to solicit that advice.

01:35:15.060 | And what my coach actually said,

01:35:16.620 | he was right about certain things.

01:35:18.020 | There are certain areas where I was under indexed on.

01:35:20.220 | Like I was not, you know,

01:35:21.540 | spending nearly enough time on my bike, for example.

01:35:23.540 | Like after that, I was on my bike six hours a day

01:35:27.900 | on the weekends.

01:35:28.740 | That's a lot of time to spend in the saddle,

01:35:30.060 | just like, just kind of, you know,

01:35:31.300 | and that was like, you know,

01:35:32.140 | for a couple of months leading up to it.

01:35:33.540 | He was right on certain aspects of it.

01:35:35.860 | And, but I kind of had to look internally and go, okay,

01:35:37.420 | like, what is he kind of missing about who I am

01:35:40.260 | and like what I kind of know I'm capable of at this point.

01:35:43.140 | I mean, it was a nail biter.

01:35:44.060 | I mean, going into the thing, you know, it's, you get,

01:35:45.860 | and this is the same thing with launching Bolt.

01:35:47.260 | It's like, or launching anything, launch day, race day,

01:35:50.980 | you kind of go in, you're like, all right, here we go.

01:35:53.820 | Like, we're going to find out,

01:35:55.060 | we're going to find out, you know,

01:35:56.140 | how based in reality I was about all the decisions

01:35:59.140 | that led to this moment.

01:36:00.180 | And so he was going and doing the Ironman

01:36:01.860 | in like six months.

01:36:02.700 | Most people spend, you know, the folks he trains,

01:36:05.060 | usually it's, you know, one to two years on this stuff

01:36:07.900 | before you do try and do a full.

01:36:09.620 | You know, it's like going and kind of doing it

01:36:10.860 | in that sort of timeframe.

01:36:11.820 | It's very similar to the same sort of skill set

01:36:15.180 | of going and building products.

01:36:17.020 | You have to really kind of look at the base reality

01:36:19.780 | and go make your own assessment on it, right?

01:36:22.220 | So.

01:36:23.180 | - Cool.

01:36:24.020 | - Great story to wrap.

01:36:25.100 | Thank you so much, Harry.

01:36:26.180 | - Thanks. - Thanks for your time.

01:36:27.100 | - Heck yeah.

01:36:28.580 | (upbeat music)

01:36:31.180 | (upbeat music)

01:36:33.780 | (upbeat music)

01:36:36.500 | (upbeat music)

01:36:39.100 | [BLANK_AUDIO]

0 to over $8M ARR in 2 months as a Claude Wrapper (Bolt.new, Qodo)

Chapters