Survive the AI Knife Fight: Building Products That Win — Brian Balfour, Reforge
00:00:00.000 | All right, I need everybody to take a deep breath here because I'm about to stress you
00:00:20.320 | out. But hopefully at the end, I'll relieve that stress a little bit with some ideas and
00:00:25.360 | solutions for you. So I need everybody to just think for a second. Reflect on the past 45 days
00:00:30.960 | and think about all the possible things that have gone on in our industry and all the product
00:00:35.440 | launches. Let me highlight just a few for you. Notion launched a Granola, Glean, and ChatGPT
00:00:41.140 | competitors. Figma launched a Canva, Framer, Illustrator, and Lovable competitor. Atlassian
00:00:45.760 | launched a Granola, Glean competitor plus Claude integrations. Anthropic launches a Glean competitor
00:00:50.460 | with Claude integrations. Google launches Codex, Lovable, and many other competitors. OpenAI
00:00:55.260 | bought a Cursor competitor, launches Codex, and a lot more. Right? This is just one little
00:01:02.400 | microcosm of the entire tech industry. But if you look around at all the different categories
00:01:08.080 | of software right now, the same exact thing is happening. And I haven't even mentioned the
00:01:13.880 | horde of startups, well-funded startups, that are getting funded in every single one of these
00:01:19.000 | spaces as well. And among all of this chaos, we have companies that are essentially collapsing
00:01:25.140 | in months rather than years. Chegg was one of the first ones to go that declined over 90%
00:01:30.500 | in the matter of months. And of course, Stack Overflow was one of the early victims as well
00:01:35.140 | when ChatGPT launched. So this gets to the number one question that we all need to be answering.
00:01:41.060 | Right? A lot of people at this conference are talking about how product is doing more engineering,
00:01:46.000 | engineering is doing more product work, designs doing more product work, all the tactical, all the
00:01:50.060 | technical, all of those different infrastructure. But none of that matters. None of it matters unless
00:01:55.220 | you answer this question. What do I build and why will it win? And the interesting thing about this
00:02:00.960 | is this was always the job of product. It just happens to be that over the years, it got marred in all of
00:02:08.160 | this project management, agile process, all of this type of stuff. But this is what always separated
00:02:14.580 | great product managers from good product managers and product leaders. This is Sean Clouse. He's the
00:02:20.640 | chief product officer at Confluent. He was formerly chief product officer at MuleSoft. He was the first
00:02:25.400 | head of growth at Atlassian as well. And I thought he encapsulated well. He said, you're constantly trying to
00:02:30.380 | get ahead. You're trying to find the angle. The question that has not yet been asked that gives you an
00:02:34.640 | insight that is not being actioned by other people. It doesn't just have to be an insight. It has to be
00:02:39.580 | an insight that others are not actioning. Because if you find that insight and others aren't actioning it,
00:02:43.940 | that is your competitive advantage. Now, the problem is, is that this question has gotten 10x harder.
00:02:51.600 | This is a rough map of Gettysburg. And I thought it was a good analogy because this was one of the
00:02:56.820 | bloodiest battles in the Civil War. And this kind of represents the map that we are all playing in in the
00:03:02.480 | competitive environment right now. We have fast, huge moving incumbents like Microsoft, Google,
00:03:08.020 | and Meta. There are these new huge horizontal platforms like ChatGPT and Anthropic that are
00:03:13.340 | eating up major use cases. We have foundational shifts in the technology landscape, not on a yearly
00:03:19.280 | basis, on a monthly basis. And there are hordes and hordes of startups being funded, including five or
00:03:26.080 | six in every single category that has traction by YC, every single cohort.
00:03:30.560 | This is you sitting in the middle of all of this, right? And the question is, is how in the world do you find a
00:03:40.040 | seam among all of these players to potentially find some traction and win? That's the question we have to answer
00:03:47.520 | before any of the other stuff like technology, infrastructure, or even what our roles are
00:03:53.000 | in the organization. I'm Brian. I'm founder and CEO of Reforge. And if you notice, I have a little bit
00:03:59.380 | more gray hair and wrinkles from this picture because I've been around in tech for about 25 years, been doing
00:04:04.220 | startups the whole time. I played in some pretty competitive environments. I helped HubSpot launch their CRM
00:04:10.440 | almost a decade ago. And at that time, that was a crazy competitive category. People thought we were
00:04:17.140 | bonkers. My guess is if I took a raise of hand, probably over 50% of your companies are using that CRM
00:04:22.340 | today. Now, that was a competitive environment. But what I'm experiencing now and what we're all
00:04:26.820 | experiencing is probably 10x that. And so a little history about Reforge is that we've been around for
00:04:33.600 | about 10 years. We've helped thousands of product teams, including all the ones you see here, over 100,000
00:04:38.180 | professionals. I hope some of you have been part of Reforge in the past. And the way that we've done it
00:04:42.700 | is that we've built a community of over 400 experts on the front lines to decode all of their best
00:04:47.700 | practices. We started by doing that with 40-plus expert-led courses, including our AI courses. But a
00:04:53.980 | couple years ago, we started to take a shift and started to encode all of this knowledge into AI
00:04:58.380 | agents. Our first one, Reforge Insights, which acts like your AI product researcher. Our second one,
00:05:03.900 | called Compass, is your AI project manager, that takes care of all of those low-level,
00:05:08.320 | low-value tasks that involve product management automated for you. We have two more coming later
00:05:13.240 | this year. But back to this question. How do you win in the intense environment in the history of
00:05:19.380 | technology? I spent a few months with Ravi Mehta thinking about this exact question. He created our
00:05:24.780 | AI strategy course. He was the former chief product officer at Tinder. He also was a product leader at
00:05:29.900 | Facebook, Microsoft, TripAdvisor, and a bunch more. And the way that we start to answer this question
00:05:35.040 | is actually, we need to think about the traps. And the two most common traps are, of course, one,
00:05:39.580 | how do you, like reinventing the AI wheel? You do not need to build custom models in infrastructure
00:05:46.640 | in order to answer this question. And on the opposite side is the other trap, which is just
00:05:52.980 | implementing, copying, and pasting basic AI features like chatbots into your product.
00:05:57.900 | The answer actually lies in the middle, which is treating AI like a series of Lego blocks where you
00:06:04.900 | assemble differentiated AI features and products by integrating the best available AI capabilities
00:06:11.440 | with your product's data and functionality. Your competitive advantage will come from what is
00:06:17.020 | uniquely yours. These three things, your data, your functionality, and your understanding of unmet customer
00:06:24.360 | needs, not the AI itself. So let's think about the anatomy of a winning AI product. What are the major
00:06:31.580 | building blocks? What are the major Lego pieces? And how do you stack them together, connect them to create
00:06:37.280 | something differentiated? Well, we can start to talk about this, the AI capabilities, because there's a ton of
00:06:43.520 | Lego pieces that are emerging every year, whether it's the pre-trained AI models, or the
00:06:48.520 | abilities to perform tasks, audio processing, imaging process, all of these new capabilities that feel magical now that
00:06:55.520 | we couldn't do before. But the thing about all of these Lego blocks is you just don't have access to them, everybody
00:07:02.360 | else has access to them as well. So even though AI products and features, of course, use one of these Legos as its core Lego
00:07:10.320 | blocks, this is not where differentiation and competitive advantage comes from. That starts with one of these pieces,
00:07:17.320 | your data. Because your data is what provides context to an AI model to generate a unique output. The more unique
00:07:25.320 | your data is, the more unique output you can generate for your customer. And there's a bunch of different types of data.
00:07:31.320 | There's real-time data that the models might not have incorporated into their training set. There's user-specific data. There's
00:07:38.320 | domain-specific data, like we've seen emerging in legal, in healthcare. There's human judgment data around curation, as well as
00:07:46.320 | data. Now the question about data is, how do you actually combine multiple categories of data together to form some
00:07:53.320 | uniqueness? As well as, it's not about the quantity of your data, it's about marginal value of your data over everybody
00:08:01.320 | else, especially the big models. So how much additional value does your data add over what is already trained in the models?
00:08:10.320 | The third piece is your functionality, because this determines how the AI behaves and it gives your AI product superpowers.
00:08:17.320 | There's multiple types of Lego blocks around your functionality. Specialized workflows, unique algorithms, business
00:08:23.320 | rules, integrations, whatever it is that's baked into your product. Now the key about assembling all these pieces is that
00:08:29.320 | they work like a system. And you have to connect the system in order to build that competitive differentiation.
00:08:36.320 | Let's start with this. Your data is what provides and informs the AI's understanding. It's what helps the AI generate a unique output.
00:08:46.320 | And that unique output, as a result, is what helps you build an additional repository of unique data so that it continues to flow in a flywheel.
00:08:58.320 | On the other side of the spectrum is your functionality. Your functionality in your product is how your product controls the AI
00:09:05.320 | actions, how it interacts with AI, when it calls it to create a delightful user experience. And in addition, AI is increasingly able to call
00:09:15.320 | tools in the functionality of your product itself. And those two things work together as a system as well.
00:09:22.320 | So, let's take all of this theory and let's put it into practice. Let's talk about a product, granola.
00:09:28.320 | Just by a raise of hands, how many people have either tried or used granola today?
00:09:32.320 | Okay, pretty decent amount. That's probably like 40% of the room. A year ago, that would have been zero. And I think this is an interesting case, because they entered a space that already had a horde of
00:09:44.320 | a horde of other AI note-takers, whether that was Fathom, Otter, Fireflies. There was a ton of them.
00:09:51.320 | But somehow, they found a seam. And they've garnered 40% of your attention in this room and about $50 million in funding.
00:10:00.320 | So, let's go back to those three fundamental questions in those Lego bricks. What was uniquely theirs? Their data, their functionality, and their understanding of their unmet customer need.
00:10:08.320 | So, I'm going to start with the last one. So, at the time when they entered the market space, this is just a sample of people who are already in
00:10:15.320 | market, including all of the incumbents like Zoom and Meet that have AI-native note-taking capabilities. But they were all approaching it from the
00:10:23.320 | perspective of the product is going to do something for the user. It's going to replace the full job. They want somebody else to take my meeting notes.
00:10:32.320 | What they realized is actually there's a whole other set of customer needs that have been unmet, which is, I don't want you to take all of my notes.
00:10:41.320 | I just want you to help me take better notes, empower me around this specific task and user. And that's what they built the product around.
00:10:48.320 | Now, in order to start, they used off-the-shelf capabilities. No unique models, no custom training, nothing.
00:10:55.320 | They used Deepgram for transcription. They used Anthropic and OpenAI for some of their other functionality.
00:11:00.320 | But the uniqueness came in how they assembled the Lego blocks, starting on the left-hand side with Granola's data, right?
00:11:08.320 | Their context includes both the notes that you take as well as the transcription that they generate.
00:11:14.320 | They used the AI Lego block to generate a unique output, which is they enhance better notes.
00:11:19.320 | Those notes, over time, form a repository that starts to enable all sorts of other features that they've layered on,
00:11:26.320 | like chatting across meetings, their project workspaces, all their downflow actions.
00:11:32.320 | So they have this nice flywheel of unique context in data that's starting to spin.
00:11:37.320 | That was partially enabled by the right-hand side of the Lego blocks, their functionality.
00:11:41.320 | They used a Mac app so that they could detect when meetings started to access the system sound for transcription,
00:11:47.320 | like being right there at the user moment that they needed it to enable the AI to do those things.
00:11:54.320 | And they've also plugged into other tools and integrations like the calendar to get metadata about the meetings,
00:12:00.320 | such as attendees. So they assembled these Lego blocks to meet in that unique way to meet that unique customer need.
00:12:08.320 | Now the question is, is Granola going to survive?
00:12:12.320 | I've got no idea, right?
00:12:14.320 | It's incredibly competitive landscape.
00:12:16.320 | Because the realization is that you can't stop here.
00:12:19.320 | You can't stop by just assembling your initial set of Lego bricks.
00:12:22.320 | You have to sequence over and over again.
00:12:25.320 | You have to take those first three Lego bricks, leverage them into another unique set that you assemble.
00:12:31.320 | And you see Granola doing this.
00:12:33.320 | Now that they've enabled this, they've started to create project and team workspaces and start to enable a new set of unique use cases off of the initial layer that they did.
00:12:44.320 | They've started to integrate downstream actions like connecting to your CRM and HubSpot.
00:12:49.320 | I just saw them the other day experimenting with a company wiki that auto-updates itself.
00:12:55.320 | So they continue to sequence these things into a unique set of building blocks.
00:12:59.320 | The question is, will they keep up?
00:13:01.320 | I don't know.
00:13:02.320 | Jamin Ball, a partner at Altimeter Capital, recently wrote a newsletter and he said, the real moat is just a sequence of smaller moats stacked together.
00:13:10.320 | Each one buys time.
00:13:12.320 | What you do with that time, how fast you execute, how quickly you evolve, determines whether you stay ahead.
00:13:18.320 | If the moat used to be six to 12 months, today, it's two to three weeks.
00:13:23.320 | So, to recap, to win an AI, besides being stressed out, right, is to answer, what are your unmet customer problems?
00:13:32.320 | That's always been a part of product, right?
00:13:34.320 | The second is, what AI capabilities can solve those problems in novel ways?
00:13:38.320 | What proprietary data can power those solutions?
00:13:41.320 | And then what superpowers can our product give to AI?
00:13:44.320 | How do you assemble those three foundational Lego blocks?
00:13:48.320 | All right, thank you.
00:13:50.320 | If you're an AI engineer, we are hiring.
00:13:53.320 | Our team will be outside.
00:13:54.320 | We can play with products with instant distribution to 300,000 people.
00:13:58.320 | And if you need help with anything else, just check out reforge.com.
00:14:01.320 | Good luck.
00:14:02.320 | .