Building AI-native: Inside the stacks powering Cognition, Gamma, and Harvey

1. 0:00 – 1:39

   Intro

   1. SPSpeaker0:00

      [upbeat music]

2. 1:39 – 2:45

   Panel setup: Building at the frontier with Gamma, Cognition, and Harvey

   1. SPSpeaker1:39

      Please welcome to the stage Head of Product for AI at Gamma, Deanie Fatiha; Co-founder of Cognition, Walden Yan; Head of Applied Research at Harvey, Niko Grupin; and Head of Startups and Venture Partnerships of Anthropic, Beth Robertson. [audience cheering]

   2. SPSpeaker1:59

      [audience applauding] Hello, everyone. My name is Beth Robertson, and I lead the startup team here at Anthropic, and I am so excited to have all of you in the room with us today. Now, everyone in this room is working through the same architectural questions, uh, that come with building frontier AI companies. And these three humans next to me are absolutely no strangers to them, so I'm excited today to take us on a tour of how the bets that they've made and kind of some of the nuances they've had to navigate as they are building at the frontier. So before we kick things off, I'd love to invite us to just go down the line. Who are you? Introduce yourself, please. What's your day job? Uh, what does your business do? And what is one core bet that you took when founding your company?

   3. SPSpeaker2:44

      Core bet. All right.

   4. SPSpeaker2:45

      Yeah.

3. 2:45 – 3:39

   Founding bets: Harvey’s wager on rapidly improving models for legal work

   1. SPSpeaker2:45

      Well, first of all, Beth and Anthropic, thanks for having us. This is an incredible event. Uh, I'm Niko Grupin. I lead applied research at Harvey. Uh, Harvey is the generative AI platform for legal and professional services. Um, and bet, yeah, I mean, I think Harvey, and, and I would argue this is the case for most application layer companies, is really a large bet that model and model capabilities are gonna improve really rapidly, and that those capabilities are gonna generalize well to the legal vertical. Right? Like, a lot of people don't know this. When I joined Harvey, we were living and working out of an Airbnb. Uh, Gabe and Winston were using what I think we would call, like, small models now to, to essentially answer personal legal questions on Reddit.

   2. SPSpeaker3:29

      Wow.

   3. SPSpeaker3:29

      Right? And so it's really this wave of this, this, uh, exponential progress at the model layer that's allowed us to kind of raise the ceiling on our ambition as a company.

   4. SPSpeaker3:38

      Amazing.

4. 3:39 – 5:21

   Founding bets: Cognition’s bet on autonomous coding agents

   1. SPSpeaker3:39

      Yeah. Again, also thank you for, for having me. I'm Walden. I'm one of the co-founders of Cognition, and we build AI coding products like Devin and Windsurf. And I think the key bet that I think of when it comes to Devin was really this bet on autonomous agents.

   2. SPSpeaker3:55

      Mm.

   3. SPSpeaker3:56

      And this was even before we had agents to start with. So in many ways, our product didn't work when we first came up with it, but the vision of having something that didn't just write the code for you, but then actually had its own computer and would run the code and then actually pull up a desktop and test it and tell you when it isn't working and fix its issues and give you a finally working-

   4. SPSpeaker4:16

      Yeah

   5. SPSpeaker4:16

      ... PR at the end of it, um, very much was not possible with the set of models we had two years ago. But lots of incremental things changed and improved since then.

   6. SPSpeaker4:25

      Hmm.

   7. SPSpeaker4:25

      Um, I think around this time last year, you saw a lot of natively agentic models like Sonnet 3.5, 3.6 come out. That helped a lot. Computer use testing capabilities are becoming more common nowadays. And I think especially with the recent models we're seeing, there's a large h- long-horizon autonomy that's becoming-

   8. SPSpeaker4:43

      Yeah. For sure

   9. SPSpeaker4:44

      ... newly possible, where you can have it run on hours end, and you start to feel bottlenecked by trying to run too many agents locally. And so we see an explosion in cloud agent usage this year. I think one crazy stat we've seen as a result of new model capabilities is that our best week of twenty twenty-five, like the, the, the end of twenty twenty-five, the amount of agent usage has grown five to seven X in our customers-

   10. SPSpeaker5:07

       Wow

   11. SPSpeaker5:08

       ... just so far in twenty twenty-six. So it looks like it's gonna be an explosive year for cloud agents, and it's-- the model capabilities are only gonna keep growing from here.

   12. SPSpeaker5:16

       Amazing. Well, we'll get back to your current bets that you're taking in a minute, but let's go down to you, Deanie.

5. 5:21 – 6:32

   Founding bets: Gamma’s bet on removing the “90% formatting work” in visual communication

   1. SPSpeaker5:21

      Well, excited to be here. Um, first of all, uh, especially excited to be here with all of you folks. We're big Devin users. We're big Claude users.

   2. SPSpeaker5:27

      Yeah.

   3. SPSpeaker5:28

      We're big Gamma users. And Niko, we're gonna have to figure out how to get big on Harvey too. [laughs]

   4. SPSpeaker5:31

      Get you on Harvey. Come on. Where's your GC?

   5. SPSpeaker5:33

      [laughs] Um, hey, everyone, I'm Deanie Fatiha. I'm head of AI product at Gamma. Uh, Gamma is a visual communication platform for professionals. We have seventy million users and growing. Um, Gamma started with the simple observation that people like us, when we're, you know, communicating our highest stakes ideas, we're usually doing it through a visual artifact, you know. Presentations to your investors, uh, proposals to your customers, marketing sites, social posts to spread your ideas. Um, and what do we do when we have to communicate those ideas? We spend, like, ten percent of our time thinking about the core insight and ninety percent of the time on the design and the formatting and the futzing of the details. Um, Gamma's big bet was that AI-- with, with AI, that we could take away that ninety percent of the futzing that people spend their time on. You as a professional, uh, bring the core insight that only you are capable of bringing. Gamma will take it, flesh out the idea, structure the narrative, design it, and make it look beautiful so that your ideas can be cast in the best light possible.

6. 6:32 – 8:03

   What made these products possible: key ecosystem shifts and capability inflections

   1. SPSpeaker6:32

      Yes, and it's a beautiful experience. So thank you all for building what you've built. We love it. Um, I wanna just go back to the origin story before we kind of get to the present. Um, there was presumably a window where you just had an idea, and it was finally made possible.Talk to me about what was happening kind of in the, the ecosystem. What shifted that made it first possible for the first time for your product or service to exist? Was there a moment? What was kind of the, the moment?

   2. SPSpeaker7:03

      I'm happy to, I'm happy to take the first stab at this. Um, and to be honest with you, I, I do think there was a first moment, but I think honestly with every big wave in AI progress, Gamma's had tailwinds that have, you know, also evolved our product. So, uh, Gamma actually started in 2020 before the recent AI wave really had hit and landed. Um, but when image models started to get come out and get good, and when LLM instruction tuning started changing how we interact with LLMs and what we can get out of them, that's when Gamma really had an aha moment, and that's, that's the inflection point that birthed Gamma as we know it today. Um, but over time, I mean, uh, again, like I said, tides of AI progress have evolved how we, we see and build our product. Uh, the next big thing that I think we did was when LLM tool calling started getting really good. We jumped on it. We built our first agentic experiences. Uh, to this day, editing using our agent is one of our biggest differentiators, and that happened because of the, you know, big wave in tool- LLM tool calling and, uh, agent, uh, uh, or- uh, orchestration.

7. 8:03 – 9:40

   Gamma’s connector strategy: MCP as workflow embedding + acquisition channel

   1. SPSpeaker8:03

      Uh, a, a big one for us as well was the MCP wave. Again-

   2. SPSpeaker8:07

      Mm-hmm

   3. SPSpeaker8:07

      ... we leaned into that early. We leaned, leaned into it heavy. It allowed us to build connectors into several other platforms. Actually, I think we built our first connector into Claude.

   4. SPSpeaker8:17

      Oh, nice.

   5. SPSpeaker8:17

      And it really changed how we think about distribution, not just our product, but our GTM.

   6. SPSpeaker8:23

      Hmm.

   7. SPSpeaker8:23

      Because all of a sudden, what that allowed us to do was we had Gamma as an agent in other surfaces and tools that our target users loved and were already living in.

   8. SPSpeaker8:31

      Uh-huh.

   9. SPSpeaker8:32

      And so what we started seeing is that not only were our existing users now able to use Gamma much more frequently and regularly because they didn't have to break their workflow and go to Gamma to continue their work, they were just doing it already in Claude, and just hitting, like, "Yes, please," and make, make a presentation out of this. Not just that, but also for new users, it was a huge acquisition channel for us.

   10. SPSpeaker8:52

       Hmm.

   11. SPSpeaker8:53

       Um, users started discovering us through the, uh, platforms like Claude because that's where they were already working and living. So that was another sort of, I would say, evolution that happened because of MCP and, um, honestly, we're, we're thinking about the next evolution of our product already now with communication being agent mediated. We're asking ourselves, "What does Gamma look like in this next evolution?" And that's gonna be something that, you know, is I think, uh, something that we spend time on this year as well.

   12. SPSpeaker9:17

       Love it. So is the MCP acquisition channel your primary way that you're getting customers today?

   13. SPSpeaker9:22

       Sorry?

   14. SPSpeaker9:22

       Is the MCP kind of ac- acquisition channel your primary, like, product-led growth motion today?

   15. SPSpeaker9:27

       Oh, uh, it's, it's, it's a huge one. Uh, I don't know if it's my primary. And we also have our, uh, like because of MCP, we have connectors now not just in Claude, but in several other B2B tools, so it's becoming a really good acquisition channel, yeah.

   16. SPSpeaker9:36

       Fascinating. How about you, Walden, for Cognition?

8. 9:40 – 11:14

   Cognition’s agent engineering lessons: deleting early systems and relying more on file systems

   1. SPSpeaker9:40

      Yeah. One of, um, one of the things I think you learn when you're building agents starting in like 2024, you build a lot of things that you go delete. And so o- one of the funniest examples of this, models back in the day, they didn't know how to edit code. They could only spit out entirely new files of code.

   2. SPSpeaker9:57

      Yeah.

   3. SPSpeaker9:57

      And so you, you had to go to really creative measures to get these things to edit code. And i- if you look into it, it's actually really interesting tech that companies like Cognition worked on to make this happen using like a combination of speculative decoding and inference techniques. Um, but this all changed once models were actually natively RL'd to be able to edit code, grab code on their own, and so that was a huge leap for us. I think another one that people underestimate is just like how natively these things use file systems nowadays.

   4. SPSpeaker10:26

      Mm-hmm.

   5. SPSpeaker10:27

      Like before we, we had a whole custom planning system, the ability to like cr- create these long horizon trajectories and follow the steps cleanly, and now you can kind of just tell the agent, like, "Write your plan down and follow it," and it'll know how to like look at the file system and figure it out. Same with memory. I think there's, uh, probably a lot of you in the audience right now thinking about how do you make memory for your agents, and you had to build a lot of very custom systems before. Now you can use like your file system in a much deeper way. People moving from RAG to file systems. So, um, I think that's like probably a pretty actionable thing that we've found is like h- how you should like build your models and agents now that that capability's improved a lot, and it's unlocked a lot of long horizon work you can do with Devin.

   6. SPSpeaker11:08

      Hmm. I love that. So autonomous agentic work-ish.

   7. SPSpeaker11:12

      Yeah.

   8. SPSpeaker11:12

      Love it. Let's go over to you, Niko.

9. 11:14 – 13:03

   Harvey’s multi-agent future: mapping law-firm hierarchies onto agent hierarchies

   1. SPSpeaker11:14

      Sure. Yeah. No, I think actually I love the phrase Walden used during the intros, which is that model capabilities make things that were impossible possible. So for Harvey, um, I think there are really three inflection points. The first is just the emergence of foundation models, so scale, um, lead- leading to kind of emergent reasoning capabilities that brought us from this like Reddit legal Q&A world to solving big law legal work.

   2. SPSpeaker11:38

      Hmm.

   3. SPSpeaker11:38

      Um, second, of course, reasoning models kind of late 2024, early 2025. That was really where we started working on what I would call like workflow automation.

   4. SPSpeaker11:50

      Yeah.

   5. SPSpeaker11:50

      So with enough kinda elbow grease, enough kinda predetermined model calls, retrieval steps, search steps, you can really solve any individual task. And then this past winter, I think Opus 4.5 was really the tip of the spear here with coding agents. You just offload planning and orchestration to these models.

   6. SPSpeaker12:07

      Hmm.

   7. SPSpeaker12:07

      And I think the coolest part about it for us at Harvey is [lip smacks] not just what happens at the individual agent level, but with multi-agent coordination.

   8. SPSpeaker12:16

      Okay.

   9. SPSpeaker12:17

      So work in, in, in law firms is very hierarchical, right? You have a partner who's working with, you know, Dario or, or your GC, uh, on a months-long project. They're gonna decompose that into a number of tasks they give to their senior associates who are gonna complete those tasks over weeks, who are gonna break those down into tasks that they can give junior associates to complete on the order of days.

   10. SPSpeaker12:37

       Mm-hmm. Yep.

   11. SPSpeaker12:38

       And now with models like Opus 4.7 and increasingly infrastructure to orchestrate agents like CMA, I don't know if Jess and team are, are in the audience right now, but it's been awesome working with them sort of onboard to this managed agents infrastructure.

   12. SPSpeaker12:53

       Love it.

   13. SPSpeaker12:53

       You can model agen- uh, agentic systems in the same kinda hierarchical fashion, which unlocks-Massive potential for our platform

   14. SPSpeaker12:59

       So you're able to out- outsource a bunch of tasks to these agents.

   15. SPSpeaker13:02

       Yeah, exactly.

10. 13:03 – 14:40

    Hard-earned lesson: every major model wave can force a full re-architecture

    1. SPSpeaker13:03

       That's beautiful. Um, thank you for sharing kind of like the early bets. I think now just looking backwards at those bets that you made, for better or worse, I think we all in hindsight, you know, may, may have changed some things. What would those have been for your, for your early bets, and, um, how... what did you learn the hard way that kinda make that true for you today as a word of caution to anybody who's kinda setting this up?

    2. SPSpeaker13:26

       Uh, I'm happy to start here because I just mentioned these three inflection points.

    3. SPSpeaker13:30

       Yeah.

    4. SPSpeaker13:30

       What I didn't mention is that for each of those three inflection points we had to completely re-architect our product.

    5. SPSpeaker13:36

       Oh, wow.

    6. SPSpeaker13:36

       Right? Um, and so I think the har- the, the kind of like hard-earned, uh, lesson here is you can't make point-in-time decisions and then stick to them. You almost have to like even, like e- this is kind of a new playbook for engineering where you kinda wanna cut things once and then just scale it, right?

    7. SPSpeaker13:56

       Yeah.

    8. SPSpeaker13:56

       You have to be able to project forward progress and, and in many cases, like exponential progress.

    9. SPSpeaker14:03

       Mm-hmm.

    10. SPSpeaker14:03

        Um, and so I think the coding agent example is a great one where six months ago if you asked me what our architecture looks like, it's fundamentally different than what it, what it looks like today. Uh, and if we hadn't been willing to say like, "Hey, we need to, we need to scrap this and go agent native," like we just wouldn't have these capabilities in our platform.

    11. SPSpeaker14:19

        Yeah. The ground is literally-

    12. SPSpeaker14:20

        Yeah

    13. SPSpeaker14:20

        ... shifting under us. Is, is your planning timeline truncated then? Is it like months or weeks, or how do you kinda think about that?

    14. SPSpeaker14:26

        So we still plan on a quarterly basis-

    15. SPSpeaker14:28

        Yeah

    16. SPSpeaker14:28

        ... but we have baked into sort of like the week-to-week execution just like retros.

    17. SPSpeaker14:33

        Got it.

    18. SPSpeaker14:33

        Say like, "Hey, do we need to just totally deprioritize this and prioritize something-

    19. SPSpeaker14:37

        Builds well

    20. SPSpeaker14:37

        ... in its favor?" Yeah.

    21. SPSpeaker14:38

        Sure.

    22. SPSpeaker14:38

        It's like you need to be able to do that on a week-by-week basis.

11. 14:40 – 16:01

    Cognition’s caution: invest in observability, replay, and evals to evolve with models

    1. SPSpeaker14:40

       Yeah. Uh, I, I totally agree. That's kinda like the way of life of like building an AI right now, is you have to accept that the thing you build today is like very likely going to be scrapped in like six months to a year.

    2. SPSpeaker14:50

       Yeah.

    3. SPSpeaker14:50

       Especially if you're, you've- if you're keeping up with the times and doing your things correctly. And so I don't look back and like regret the fact that we did any of like the previous things we did and had to get rid of them. I think the actual important part that was really important to invest in, and we should've invested more in if anything, was building everything you needed to actually make sure you can keep changing your product as the quality of the model changed. And so that actually looks like the underlying logging, observability, the ability to know, like for your engineers to be able to dig into a- any decision your agent makes, replay that, figure out do new models make that decision better-

    4. SPSpeaker15:25

       Yes

    5. SPSpeaker15:25

       ... have evals for those things. Um, the easier it is for you to answer these questions about your system and why something went wrong, the easier it is for you to say, "Okay, new model capabilities mean that we have to do this differently now. We have to, um, re-architect this." And, uh, I've seen a lot of companies that kind of like build in the dark. They add a prompt, and they're like, "Oh, hopefully this like makes things better."

    6. SPSpeaker15:44

       [laughs]

    7. SPSpeaker15:45

       Um, but you really don't know. And, and so if you want to reliably work across like, you know, new model upgrades-

    8. SPSpeaker15:50

       Yeah

    9. SPSpeaker15:50

       ... different models, uh, the observability, the, your internal tooling, it's super, super important to get right.

    10. SPSpeaker15:55

        Got it. So observability, don't lock into anything for too long 'cause everything changes. How about you, Demi?

12. 16:01 – 19:06

    Gamma’s current rebuild: rebalancing speed vs ‘nailed it’ quality across workflows

    1. SPSpeaker16:01

       I mean, echoing a lot of what Nico and Waldan are saying, we're actually living through this right now where we're re- um, re-architectu- uh, re-orchestrating our entire generation architecture. Um, and this is because, you know, uh, if I take a step back in time, when we first built our generation sort of system in Gamma, uh, the breakthrough was speed, right? You come to Gamma, you put in a prompt, you s- choose some settings, you hit generate, and you go from a blank page to a beautifully designed deck in a minute. And I remember the first time I experienced that as a Gamma user, and I thought it was mind-blowing. Like that was the breakthrough, right?

    2. SPSpeaker16:36

       Yes.

    3. SPSpeaker16:36

       Uh, when AI, you know, again, the, the, a recent wave of AI hit, I felt like s- feel like speed was the thing that everybody was so mesmerized by. But things have changed and evolved. Not only has, um, AI gotten much smarter in, in terms of its ability to self-critique and coach the user and push back and, you know, reason and s- research and so much more, uh, user expectations have evolved rightly along with that as well. So today, uh, given the workflow, a lot of the times users are willing to wait instead of having a, "Wow, that was fast," moment. They wanna have a, "Wow, it really nailed it," moment, and they're willing to wait for that moment.

    4. SPSpeaker17:14

       Mm.

    5. SPSpeaker17:14

       And so now we're rethinking, um, we're sort of, uh, you know, re-architecting our system to be able to, A, take advantage of the latest sort of capabilities of AI at the expense of speed in some cases.

    6. SPSpeaker17:25

       Interesting. Okay.

    7. SPSpeaker17:26

       [laughs]

    8. SPSpeaker17:26

       Because the thing that, the nuance here isn't that it's speed versus quality. It's always speed and quality. Um, because I just told you all, about all the fa- uh, uh, you know, all of the stuff about how generation people are willing to wait longer. We find when they're editing and making the tweaks, they want snappy, responsive, fast, uh, at the cost of quality sometimes.

    9. SPSpeaker17:45

       Mm.

    10. SPSpeaker17:45

        So it's one of those systems n- like we're sort of taking a step back and figuring out how do we build a system where it's not speed versus quality. These are all parameters that we can dial up and down, and often even depending on sort of the workflow, pass that choice onto the user-

    11. SPSpeaker17:59

        Mm

    12. SPSpeaker17:59

        ... so they can decide what they're in, in the mood for today and what they have time for today. So, uh, designing a system that's sort of extensible across those parameters and, uh, flexible is, is sort of, uh, the exercise that's happening at Gamma.

    13. SPSpeaker18:10

        And are you using like different models to be able to play this whole field?

    14. SPSpeaker18:14

        Totally. Uh, we, we were always using different models, but I think we were sort of al- almost optimizing them for a certain outcome. And now what we're trying to do is, well, we have a, a, a, a whole, uh, layer of, um, models that we're orchestrating, but then now we're also trying to sort of give the user the power to decide, "You know what? Today I need it fast. I don't have time."

    15. SPSpeaker18:33

        When you have-

    16. SPSpeaker18:33

        "I have, I have customers who are in their car who wanna be... Make me a deck. I'm on my way to my customer."

    17. SPSpeaker18:37

        Yeah.

    18. SPSpeaker18:37

        So that's when you need speed versus quality, and we wanna give that, uh, the power to the users to sort of decide.

    19. SPSpeaker18:43

        Fascinating. I love that. Uh, you brought us to the topic that I was gonna logically flow to next, which is like what is the big bet that you're making today, um, that you're kinda betting the next three months? We'll give it a shorter time horizon on. Or we can pivot it-

    20. SPSpeaker19:00

        Yeah. [laughs]

    21. SPSpeaker19:00

        [laughs]

    22. SPSpeaker19:00

        ... to what is the big bet-

    23. SPSpeaker19:01

        [laughs]

    24. SPSpeaker19:02

        ... uh, that you're making today that founder in- founders in this room should be paying attention to?

13. 19:06 – 20:41

    Big bets now: self-driving codebases and cloud agents as the new software org baseline

    1. SPSpeaker19:06

       Oh, I- I'm happy to answer the, the, the bet question as well.

    2. SPSpeaker19:09

       Yeah.

    3. SPSpeaker19:09

       Um, no, I, I, I'm very excited for, like, the, this next year. Um, it feels like cloud agents are kind of becoming very possible now over these last few months. And I think the thing that everyone kind of has really high demand for, whether you call it self-driving code bases, whether you call it the software factory, is just how can you just take everything that you do as a software org and automate as much of that as possible. And then whatever does need human to, to review, to look at, um, have the AI lift that up to human. You're going from a world where it's, like, default, like, human at the driving wheel and AI can take over to, like, the AI is the one driving the projects end to end-

    4. SPSpeaker19:46

       Mm-hmm

    5. SPSpeaker19:46

       ... doing the planning, the coding, the reviewing and testing, um, and figuring out when it actually needs to pull the human in. Um, and when you make this shift, that's, that's absolutely crazy, and it significantly increases the amount of work you can do as a software org. So I, I've been telling people today who I've been meeting, like, Cognition as a company, we're like 50 engineers right now.

    6. SPSpeaker20:06

       Yeah.

    7. SPSpeaker20:06

       Um, but every engineer has like 10 Devins that they're using to do everything they need to do, and the role of the software engineer has to change when your code base becomes self-driving and you have a self-maintaining, uh, code base. There's a lot of big companies in the world who have to make this change over the next three years, and also a lot of what we're doing is going in and partnering with them to help them figure out how do you restructure the way you think about coding, you think about project management to get to this point. Um, but there's a lot of really cool automations and setups you can do to actually get to you, to this frontier. It's what we're hearing from all our customers, and that's, that's what we're building our product towards right now.

    8. SPSpeaker20:41

       Mm-hmm.

14. 20:41 – 23:18

    Harvey’s bet: collaboration + infrastructure constraints to turn personal AI into org productivity

    1. SPSpeaker20:41

       Yeah. I mean, I definitely agree with this transition from, like, reactive intelligence to proactive intelligence.

    2. SPSpeaker20:47

       Yeah.

    3. SPSpeaker20:47

       Where my brain went with this is actually more, like, back to the le- the less exciting-

    4. SPSpeaker20:51

       I saw you pause. [laughs]

    5. SPSpeaker20:51

       ... part of this. [laughs] Uh, which is individual, uh, productivity gains from AI distributed widely-

    6. SPSpeaker21:01

       Yeah

    7. SPSpeaker21:01

       ... does not equal organizational productivity gains.

    8. SPSpeaker21:05

       Mm.

    9. SPSpeaker21:05

       Right?

    10. SPSpeaker21:06

        Fair.

    11. SPSpeaker21:06

        And so this is, like, almost exactly what Walden was just describing, is, like, the nature of your role changes, is like if you move 10X faster, that means you can also move in the wrong direction-

    12. SPSpeaker21:17

        Yeah.

    13. SPSpeaker21:17

        Yeah.

    14. SPSpeaker21:17

        ... 10X faster-

    15. SPSpeaker21:18

        Yeah

    16. SPSpeaker21:18

        ... or make mistakes 10X more quickly or, or the blast radius is 10X larger.

    17. SPSpeaker21:22

        Mm.

    18. SPSpeaker21:22

        Um, so how do you actually move one layer of abstraction up in the decision ma- making process, right?

    19. SPSpeaker21:27

        Mm.

    20. SPSpeaker21:27

        So, like, the engineering equivalent here would be like is this the right architecture, is this scalable-

    21. SPSpeaker21:30

        Yeah

    22. SPSpeaker21:30

        ... is this secure, et cetera. I think we're gonna see that broadened into more general knowledge work as well. And then I think the part that I'm really excited for Harvey to play in this is collaboration, right? Like, what does the interface for that even look like?

    23. SPSpeaker21:43

        Yeah.

    24. SPSpeaker21:43

        Right? If you wanna enable lawyers to collaborate within their firm, you wanna enable lawyers to collaborate with their clients outside of their firm, and then you want humans to be able to collaborate with agents kind of all in the same workspace. Uh, I think it's a really interesting problem from all aspects, product, AI, infrastructure. Um-

    25. SPSpeaker22:02

        Mm

    26. SPSpeaker22:02

        ... and I think it's gonna be a big part of the rest of our year and, and going into next year.

    27. SPSpeaker22:06

        And knowing that, like, I know that you're planning three months ahead, how are you thinking about that for, for Harvey? And just, like, the personal productivity versus organizational productivity stands out to me.

    28. SPSpeaker22:16

        Yeah. So this actually is where I know the tendency, especially on Twitter, is just to, like, lean into the bleeding ed- like, the frontier of model intelligence and try to pull it, push it.

    29. SPSpeaker22:25

        Yeah.

    30. SPSpeaker22:26

        This is actually where we're taking a step back and focusing on infrastructure first. So to give you an example, if you have agents running around in these workspaces, there are certain kind of, like, data constraints that law firms have. This is extremely sensitive data, right?
15. 23:18 – 24:49

    Gamma’s forward-looking bets: ‘taste’ as differentiation and designing for agent-mediated communication

    1. SPSpeaker23:18

       Incredible. Do you have any thoughts on the future, like, bets that are gonna happen in the next few months?

    2. SPSpeaker23:22

       I mean, from what Nico was saying, yeah, I think there's, like, two, two sort of bets/things should- founders should be looking out that are sort of super top of mind for, for me, and it, it echoes some of what's been said. Um, well, the first one is, it's a little bit different, which is, um, I think this year we really want to... I mean, we've always sort of thought taste is the new buzzword in Silicon Valley. And guess what? At Gamma we've been agonizing over taste for years.

    3. SPSpeaker23:46

       [laughs]

    4. SPSpeaker23:47

       Um, and so we're gonna continue down that route. We're, uh, really thinking about how to increase the visual range of what you can do in terms of design in Gamma.

    5. SPSpeaker23:55

       Mm-hmm.

    6. SPSpeaker23:55

       Uh, we've always been doing it. We're gonna double down on it a lot more, and that's gonna be something that's, like, super sort of, um, critical for us in the next few months. Um, but also the, the, the thing that, uh, both Nico mentioned and I mentioned earlier, which is we are increasingly in a world where agents are everywhere, and they're helping us, and they're mediating our world, and we're thinking about this very deeply. Like, what does Gamma look like in a world where communication is agent-mediated?

    7. SPSpeaker24:21

       Mm.

    8. SPSpeaker24:21

       Where agents are supposed to be able to use your product just as humans? How do you make your product delightful and useful both for humans and agents and human agent collaboration?

    9. SPSpeaker24:30

       Yeah.

    10. SPSpeaker24:30

        Um, so that's going to be something really, um, again, that we're spending a lot of time on this year and a, and a big area for, of focus. And I think it's something that I, I haven't at least seen enough people think about, is what does your product look like? Does your product exist in a world-

    11. SPSpeaker24:44

        Mm

    12. SPSpeaker24:44

        ... uh, in its current form, in, in a world where it's all agent-mediated? And I think that's something that everybody needs to be thinking about.

16. 24:49 – 28:05

    Lightning round: personal AI wins, near-term predictions, and parting founder advice

    1. SPSpeaker24:49

       A lot to chew on. Okay, I'm looking at the timer, and we are gonna move to our lightning round. Who's excited? Um, okay. One thing that AI has solved in your personal life that's been a massive unlock.

    2. SPSpeaker25:02

       Go

    3. SPSpeaker25:02

       We're just going down the line?

    4. SPSpeaker25:03

       Five words. I mean, you guys... Popcorn?

    5. SPSpeaker25:04

       Five words.

    6. SPSpeaker25:05

       Do popcorn.

    7. SPSpeaker25:05

       All right. This is somewhat embarrassing. I've- I've offloaded my entire weekly meal planning-

    8. SPSpeaker25:10

       Stop

    9. SPSpeaker25:10

       ... and day-to-day diet to Claude Code. I'm not kidding you.

    10. SPSpeaker25:13

        Check it out.

    11. SPSpeaker25:13

        I have a log-

    12. SPSpeaker25:14

        Are you eating better?

    13. SPSpeaker25:14

        ... on my local machine, and it... What?

    14. SPSpeaker25:16

        Are you eating better than ever?

    15. SPSpeaker25:17

        Oh, yeah. I- I, like, massaged all sycophantic behavior out of the model with- [laughs]

    16. SPSpeaker25:22

        [laughs]

    17. SPSpeaker25:22

        [laughs] With the Claude.

    18. SPSpeaker25:22

        Oh.

    19. SPSpeaker25:23

        Yeah.

    20. SPSpeaker25:23

        Massive unlock. Okay, that was more than five words.

    21. SPSpeaker25:25

        Yeah, sorry.

    22. SPSpeaker25:25

        Uh, travel planning.

    23. SPSpeaker25:26

        Sorry. My bad.

    24. SPSpeaker25:27

        That was a good one.

    25. SPSpeaker25:27

        Travel planning.

    26. SPSpeaker25:28

        What was it? Travel planning?

    27. SPSpeaker25:29

        Yeah.

    28. SPSpeaker25:30

        Love it. I'm gonna give you more than five words. Um, [laughs] I- I organized a concert at my home, uh, a few weeks ago, and, uh, Claude actually was the event planner for it. It found me caterers. It took care of my decor. It advised me on the setup. It was- it was amazing. You guys are so cool.

    29. SPSpeaker25:46

        [laughs]

    30. SPSpeaker25:46

        It's like travel, food, and concerts at your house.

Episode duration: 28:15