Using agents to build an agent company: Joao Moura

00:00:00.000 | -

00:00:02.000 | 10 million, 602 thousand, 922.

00:00:18.360 | Over 10 million and a half.

00:00:24.200 | That's how many agents got executed with crew AI

00:00:27.200 | in the last 30 days.

00:00:28.440 | And this is crazy.

00:00:32.800 | I'm so impressed by this because this is real.

00:00:36.360 | It's real.

00:00:37.200 | And it's moving way faster than most of people think.

00:00:40.600 | And I assume most of you here know what AI agents are,

00:00:44.320 | but I'm gonna catch you up if you don't real quick.

00:00:46.800 | So, LLMs, we all know them, ChatGPT.

00:00:49.560 | Turns out, great to create content.

00:00:51.660 | And they're so good that they almost look they're reasonable.

00:00:54.740 | They can choose between left and right and up and down

00:00:57.000 | and all their other options.

00:00:58.480 | And if you get them to chat with themselves

00:01:01.120 | or to a copy of them, guess what?

00:01:03.560 | You can leave your room.

00:01:04.640 | You have an agent.

00:01:05.920 | Basically, it can take its own decisions,

00:01:07.760 | it can use the tools, and it can be autonomous.

00:01:10.640 | So, if you didn't know what an agent was, there you go.

00:01:14.080 | Now, you might be asking, great, now what?

00:01:17.200 | What's up to me?

00:01:18.520 | Well, we have been building automations as engineers for decades.

00:01:22.640 | And it usually starts pretty straightforward.

00:01:24.480 | It's like, hey, I want to go from A to B, but then what happens when you add C and then you add D?

00:01:31.000 | And these things can get complex pretty quick, and that's how legacies and headaches are born.

00:01:36.000 | But it turns out that with agents, you don't necessarily need to do that.

00:01:40.120 | You don't need to connect the dots.

00:01:41.600 | You give it the options, and the agents can adapt to the circumstances that they are met.

00:01:47.440 | And they can do that in real time.

00:01:49.600 | So, that allows them to build automations that were never possible before, that you couldn't do it.

00:01:55.600 | And when you think about the anatomy of these agents and what they look like, they might look pretty simple at first.

00:02:01.600 | You might say like, well, you have an LLM in the center and you have tasks and you have tools, but once they start to building these things in production for real, you quickly realize that you've got to think about, well, I need a caching layer, I need a memory layer, I need to train them, I need to find a way to add guardrails, and so much more that goes into that.

00:02:20.200 | And then now you want to like them to talk to each other, and that adds another complexity layer.

00:02:25.240 | And then, when they're in a crew, you want to still think about the caching, but now it's shared, and the memory, and now it's shared, there's so much that goes into this.

00:02:33.920 | And then you can go one extra level and get multiple crews to talk to each other.

00:02:38.240 | All that goes to say that the way that we have been building software is changing a lot, is to think about the way that we used to do, it's very strong time.

00:02:46.520 | All the software that we have built is very strong time.

00:02:49.040 | You start with knowing exactly the inputs that are coming in.

00:02:52.640 | It's a form, it's an integer, it's a string, you know what's happening, you're summing it up, you're multiplying, and then you have a very strong output,

00:03:01.320 | to the point that you can write basically any tests, because the behavior is always the same.

00:03:06.320 | But with AI agents and any AI apps, for what it's worth, everything is fuzzy.

00:03:12.080 | You don't know what's coming in.

00:03:14.160 | Yes, it's a string, but it can be a CSV, can be a RASV, can be a random joke, and then these models are basically black box, and you don't necessarily know what's coming out of it.

00:03:24.840 | And you know what?

00:03:25.840 | I love it.

00:03:26.840 | So, this is happening now.

00:03:29.520 | And I'm being serious.

00:03:31.520 | Every single day, 100,000 crews are executed.

00:03:38.520 | And I'm talking like every day.

00:03:41.200 | And, I mean, I have been talking a lot about Crew.AI, and Crew.AI is a production-ready library to build and orchestrate multi-AI agents' automations.

00:03:52.200 | And we'll talk more about that in a second.

00:03:54.640 | And I don't know exactly what, like, the book definition of a production-ready framework, but I'm pretty sure that involves running more than 10 million agents every month.

00:04:06.880 | So, I like to claim that.

00:04:09.320 | And the fact that we have been building this means that we get a lot of exposure to a lot of use cases.

00:04:15.720 | What are people building out there?

00:04:17.320 | How they're using this?

00:04:18.640 | And it's so good.

00:04:19.760 | Before I move along, really show off hands real quick.

00:04:22.600 | Who has tried Crew.AI?

00:04:24.360 | Raise your hands.

00:04:25.400 | I like this.

00:04:27.680 | But we're going to get an even higher number.

00:04:30.040 | So, I'm Joel.

00:04:32.400 | My name is kind of hard to pronounce.

00:04:33.880 | I go by Joe sometimes.

00:04:35.280 | Nice to meet you.

00:04:36.240 | I'm the CEO and founder of Crew.AI.

00:04:38.320 | And the way that I build this company has been a very interesting journey.

00:04:42.160 | Everything starts back in Brazil.

00:04:43.760 | I'm a long way from home.

00:04:45.520 | And everything started with my wife.

00:04:48.760 | I'm very blessed to have a very smart wife.

00:04:50.720 | And I have been working in Clearbit for many years before starting Crew.AI.

00:04:55.200 | And my wife told me, you know what?

00:04:57.000 | You're putting a lot of interesting stuff with like LLMs.

00:04:59.600 | You should be sharing more about it online.

00:05:01.840 | But I suck at writing LinkedIn posts.

00:05:04.640 | So, as a good engineer, I was like, hey, I'm going to write some agents to do that for me.

00:05:08.640 | And it turns out it fucking worked.

00:05:13.120 | I got so many views.

00:05:14.520 | Everyone was going into my LinkedIn and I was soaked.

00:05:17.120 | I was like, you know what?

00:05:18.080 | I want to automate my life away.

00:05:19.520 | I don't want to do anything anymore.

00:05:20.760 | I'm going to just relax.

00:05:22.080 | Well, for my surprise, it didn't turn out that way.

00:05:25.120 | And I start to build in Crew.AI because I want to use the same thing to build more and more agents.

00:05:31.200 | The problem was, all that happened in my anniversary.

00:05:35.680 | So, you can see that I was having a lot of fun.

00:05:39.280 | But at the same point, my wife doesn't like me to spend too much time in the computer when we're in the holidays.

00:05:44.480 | But little did she know that I was working super early.

00:05:47.520 | And I was hacking away.

00:05:48.880 | And I was building Crew.AI.

00:05:50.560 | And things start to cook off.

00:05:52.320 | And we start getting bugs.

00:05:53.920 | Like rabbit hole reports, hallucinations, two errors.

00:05:57.600 | I'm not going to lie, things got a little crazy for a hot second there for us.

00:06:00.800 | But it turns out, with that, I also created a GitHub community.

00:06:04.640 | Over 16,000 stars in GitHub.

00:06:07.440 | A Discord community with over 8,000 people.

00:06:11.440 | And then someone created a Reddit.

00:06:13.040 | I didn't know, but we have a Reddit.

00:06:14.560 | And that's amazing.

00:06:15.360 | And then not too long after that, engineers start to like reach out.

00:06:20.000 | And then companies start to reach out.

00:06:21.760 | And I'm talking about hundreds of thousands of people.

00:06:23.840 | And I'm like, all right, we need to scale this up.

00:06:26.320 | But then how we do it?

00:06:27.600 | How do you scale a company in such a competitive market?

00:06:31.360 | Guess what?

00:06:32.000 | The answer was in front of me all along.

00:06:34.080 | We need agents.

00:06:35.760 | And so did I build some damn agents.

00:06:40.640 | So, let me tell you how I start.

00:06:41.920 | I was like, hey, let's start simple.

00:06:43.760 | This is a company.

00:06:44.640 | What do we need?

00:06:45.280 | We need marketing.

00:06:46.400 | So, I build a marketing crew.

00:06:48.160 | I was like, I'm going to build first a content creator specialist, a social media analyst,

00:06:54.160 | a senior content writer, and a chief content officer.

00:06:57.520 | Bring them out together.

00:07:00.000 | This is my marketing crew.

00:07:01.600 | I'm going to shovel in rough ideas that kind of suck.

00:07:03.920 | And I want to get something great.

00:07:05.520 | So, they're going to check X and check LinkedIn and what other people were talking about this.

00:07:10.000 | They're going to search the internet and learn more about the topic.

00:07:12.960 | They're going to look at my previous experience and they're going to give me an incredible draft.

00:07:17.440 | And then I start posting it out.

00:07:19.360 | And again, guess what?

00:07:20.880 | It worked.

00:07:22.240 | We got 10x more views in 16 days, 16 freaking days.

00:07:27.360 | And I was loving it.

00:07:28.640 | But with that came a problem.

00:07:31.600 | I was like, well, I need to move on.

00:07:32.960 | I need to like now serve these people.

00:07:34.720 | How do I qualify them?

00:07:36.320 | Well, I need to go the next level.

00:07:39.200 | I did a simple crew.

00:07:40.240 | I'm going to step up the ladder.

00:07:41.680 | I'm going to do the next step, the higher impact, but lower risk.

00:07:46.240 | So, this is what I did.

00:07:46.960 | I did a lead qualification crew.

00:07:48.560 | So, I was right.

00:07:50.240 | I'm going to bring up a lead analyst expert.

00:07:53.360 | I'm going to bring an industry research specialist and a strategic planner.

00:07:56.880 | I'm going to wrap them together into a lead qualification crew.

00:07:59.760 | I'm going to shovel in my lead responses.

00:08:01.840 | And I want them to analyze the answers.

00:08:03.840 | I want them to compare them with my CRM data.

00:08:06.160 | I want them to research the industry and give me like a score, use cases, talking points,

00:08:11.120 | so I can jump in a meeting right away.

00:08:12.800 | Guess what?

00:08:15.600 | It worked.

00:08:16.320 | The problem is it worked too well.

00:08:18.880 | I ended up doing 150 plus customer calls in two weeks.

00:08:26.560 | It was crazy.

00:08:28.320 | You know what?

00:08:28.720 | I don't regret it.

00:08:29.360 | I love it.

00:08:30.080 | So, this is what happened.

00:08:31.360 | I start to expand more.

00:08:32.480 | Let's build more crews.

00:08:33.760 | We have marketing.

00:08:34.480 | We have lead qualification.

00:08:35.680 | Let's do code documentation.

00:08:37.200 | So, if you try crew, all those docs, we didn't write it.

00:08:40.080 | The agents do it for us.

00:08:41.600 | And I was like, I want to do more and start to do email and do more and more.

00:08:44.960 | And it works.

00:08:45.840 | So, to the point that these are some of the companies.

00:08:48.880 | They're now building with crew.

00:08:50.800 | They're using crew.

00:08:52.080 | And it's insane to me.

00:08:53.680 | And hey, if you don't like these companies, you don't believe that.

00:08:56.160 | Well, believe some of our investors.

00:08:57.680 | Darmash, CTO of HubSpot, or Jack Altman, I mean, they can vouch for us.

00:09:02.160 | We're doing pretty well.

00:09:03.200 | And what about the future?

00:09:05.680 | Like, where is this thing going?

00:09:07.600 | Well, as an LLM model, I can't...

00:09:10.560 | No, I'm kidding.

00:09:11.280 | So, actually, the DNA is not getting back in the bottom.

00:09:16.880 | This is going to be huge.

00:09:18.240 | Bigger than the internet.

00:09:19.200 | We all know it.

00:09:20.080 | Like, this is not going back.

00:09:21.520 | People are not going to stop using agents from one day to the other.

00:09:24.160 | Well, this is my advice for you.

00:09:28.880 | Be your lady adopter.

00:09:30.320 | Don't wait for other use cases.

00:09:32.240 | Start simple.

00:09:33.440 | Expand to low risk and high impact.

00:09:35.840 | That's what we did.

00:09:37.200 | But hey, I'm not going to finish here.

00:09:39.120 | In Queer.ai, we are known because we ship them fast.

00:09:42.560 | So, I'm going to make some announcements.

00:09:44.160 | Some of the stuff that we are working that I'm super excited about.

00:09:47.520 | First thing, your agents need tools, right?

00:09:50.240 | So, why don't you let them build their own tools?

00:09:53.280 | So, we are working with code execution.

00:09:57.040 | What it means that in the new version, all you got to do is create an instance of

00:10:00.400 | automated coder, command line code executor.

00:10:02.800 | Come on.

00:10:04.480 | You're not buying this.

00:10:05.440 | I'm not AutoJam or whatever other framework you're using.

00:10:08.160 | We're Queer.ai.

00:10:09.200 | All you got to do is one flag.

00:10:10.560 | Allow code execution.

00:10:11.680 | That works.

00:10:12.240 | Your agents can code now.

00:10:13.200 | You don't got to worry about this.

00:10:14.320 | Another thing that we are working on.

00:10:17.520 | You know how you do when you hire a new employee?

00:10:20.320 | You train them.

00:10:21.600 | Why not do that with your crew so you can get consistent results over time?

00:10:25.600 | Well, there's a new feature.

00:10:26.880 | Train your crew.

00:10:28.160 | It's a new CLI.

00:10:29.440 | You can run that and you can give instructions to it.

00:10:32.160 | And that's going to become baked into the memory of your agents to the point

00:10:35.440 | they're going to give you consistent results every time moving forward.

00:10:38.080 | But we're not stopping that as well.

00:10:40.880 | There's more.

00:10:41.440 | There's more.

00:10:41.440 | You know what?

00:10:42.560 | I like to think about us as we don't see agent callers.

00:10:46.640 | We want all the agents.

00:10:48.560 | So bring them all.

00:10:49.760 | We're a universal platform.

00:10:51.840 | Bring any third part agent.

00:10:53.440 | You know your Yama index agent, your link chain agent, your autogen agent.

00:10:57.840 | I mean, I don't know why you would use anything else because you've got a crew.

00:11:00.560 | But hey, come on.

00:11:01.440 | You can bring them into the party.

00:11:02.960 | And they're going to have all the crew AI agents features.

00:11:05.760 | The shared memory, the same tools.

00:11:08.000 | You're going to be able to use all of them if you want.

00:11:09.840 | And then, again, the best thing about all this, you can try it today.

00:11:15.920 | We just shipped a version before I come in the stage.

00:11:18.560 | And if you want to try it right before this call or later in today, you can give it a try.

00:11:22.240 | It's a new version.

00:11:22.960 | It's live.

00:11:23.920 | And if you, again, that is not exciting enough.

00:11:25.920 | Maybe you want to hear from another of our investors, Andrew Ying.

00:11:29.440 | So you can go and check it out.

00:11:31.600 | We put together a two-hour course on how to learn about crew AI.

00:11:35.600 | And you all got to do is go to learn.crewai.com.

00:11:38.320 | And final thing, I promise.

00:11:42.480 | I know that we're right at the time.

00:11:44.160 | Bear with me.

00:11:44.720 | This conference has been one of the best conference I have been.

00:11:48.640 | Who here agrees with that?

00:11:51.920 | Right?

00:11:52.240 | There's one thing, though.

00:11:54.480 | I heard a lot of, like, teary.

00:11:57.760 | A lot of, like, coming-sue type of deals.

00:12:01.040 | And I don't know about you, but I'm a little sick of that.

00:12:03.840 | So why don't we actually start to bring some, like, agents into production?

00:12:09.600 | So I want to talk about CREAI+.

00:12:12.480 | It's our enterprise offering.

00:12:14.400 | What are some of those companies that I showed you where you are using it?

00:12:16.800 | And with CREAI+, now you build your crews the way that they're running your terminal,

00:12:21.760 | system, but you basically can select them, push them to GitHub.

00:12:24.560 | In three minutes, they become an API.

00:12:26.480 | And I'm talking about a real API.

00:12:28.560 | I'm talking about auto-scaling, protected by a bearer token with a private VPC,

00:12:33.200 | everything that you need to run these things in production.

00:12:35.440 | And then you can also, like, one click away, export that into a React component.

00:12:40.240 | And now you basically have a UI that you can demo, and you can customize it anyway that you want.

00:12:44.560 | So you can basically connect your agents, like, in a few minutes to anything.

00:12:49.040 | And there's more.

00:12:50.080 | For the first 50 companies that sign up using this link, we're going to give you access to CREAI+ in less

00:12:56.720 | than 24 hours.

00:12:57.680 | And I also have one extra thing.

00:13:00.880 | You don't even have to build your first crew.

00:13:03.200 | Turns out, I put an all-nighter.

00:13:04.960 | I have a crew that can build your crew.

00:13:07.280 | You heard me right.

00:13:10.160 | Based on your email and your company name alone,

00:13:12.800 | this crew is going to run, is going to create your crew, push into a GitHub repository,

00:13:16.640 | and that can be the first crew that you deploy on CREAI+.

00:13:19.360 | So, hey, why don't we leave at that?

00:13:20.720 | It's starting to become a little, like, ugly for the other guys.

00:13:23.600 | So, hey, thank you so much.

00:13:24.880 | I catch you in the comments.

Using agents to build an agent company: Joao Moura

Chapters