# Move Early, Commit Carefully

Simina Pasat, Just Product 2026, 16:30. Automatic transcript of the livestream (Parakeet ASR, unedited, expect misheard names). Timestamps are positions in the talk video.

[0:01] Alright.
[0:03] Welcome back.
[0:05] Last sprint of the day.
[0:07] Although we heard people are not really doing sprints anymore.
[0:10] So last what is that?
[0:12] I don't know, last working item of the day.
[0:15] No, that doesn't really sound right either.
[0:17] Because this is definitely not a working item.
[0:19] This is gonna be super fun.
[0:21] Super fun.
[0:22] What did you say?
[0:23] Opportunity to learn.
[0:24] Opportunity to learn.
[0:25] Okay, the opportunity to learn.
[0:27] So next up is Semina.
[0:30] And so honestly, like, why is Semina here?
[0:36] So I asked Claude, so we were doing like speaker research.
[0:40] So as you do, I asked Claude and said, like, all right, I wanna something like in this and that direction.
[0:45] I can't remember what put in there, and her name popped up.
[0:48] And then I basically stalked her on LinkedIn, and I saw that she actually went to Tom.
[0:55] And I'm like, oh, that's actually interesting.
[0:57] So I wrote her on LinkedIn and uh in German, basically saying, like, I know this is a bit weird, but like, would you consider coming to Munich for the conference?
[1:05] And she replied in English, said, Look, I've been living in the US for over 10 years now, so like English comes more natural to me.
[1:11] So I'm gonna reply in English, but like short version is so we talked and like like we really loved her, and also the story she uh stories she can tell and how she can connect the stories between okay, like having worked and studied in Germany um all the way uh to the Sig uh to the Silicon Valley.
[1:29] Um that that kind of inspired us.
[1:31] She's also the speaker that had the longest way here today.
[1:35] So without further ado, all the way from Seattle, Simina Passad.
[1:48] Hello everyone!
[1:49] I'm so glad you're still here today.
[1:51] You didn't leave.
[1:52] That's great.
[1:53] You know, closing out the day, it's a little bit of a gift.
[1:57] And not because I'm the last person standing between you and your fire rubbing beer, but also because the speakers before me kind of did lots of my work for me.
[2:11] You know, we heard Martin talking about strategy and assumptions.
[2:16] And uh for Verena, they have uh no longer a backlog because AI did all their work, and I especially love Dave and uh his points of like going to look for truffles.
[2:27] Uh that was really fun.
[2:29] And uh, we also heard about Intercom twice, and it was pure coincidence.
[2:36] They didn't sponsor the conference, although maybe they should, because I'm gonna bring it up for the third time as well.
[2:43] And you know, the fact that we are talking about the same things, coming from so many different places, different countries, different industries, different continents, it's a little bit of a signal.
[2:56] So we must be on to something um right.
[2:59] My name is Simina and live at I live in Seattle.
[3:02] And most recently, I was a VP of product at GitHub, where I worked on GitHub and GitHub Copilot, and many of the things in the last eight years I've been here.
[3:12] And coming here today in Munich is a little bit of a homecoming because 15 years ago I came here for the first time to start my master's at TOM.
[3:22] And this city influenced my careers in my life in ways I couldn't have imagined.
[3:27] I even met my now husband here.
[3:30] So that was uh funny, but it was also a career launch bed for me.
[3:34] And we heard today, building is so much easier, right?
[3:39] What used to take weeks, months, years, now it's compressed in hours and days.
[3:46] So if building is so much easier, what is there left that is hard?
[3:51] Well, kind of a lot of things and everything in between.
[3:54] And I will talk a lot about a lot of these things and my journey from hybrids and tom and CDTM all the way to Xbox and GitHub in Seattle.
[4:04] But before that, I want to talk a little bit about sports.
[4:08] This year, against all expectations, I got into watching sports.
[4:14] With two small kids at home, this gave us a little bit of a family ritual.
[4:19] Something we could all relax into.
[4:21] And finally, TV, I didn't feel guilty about.
[4:25] That's great.
[4:26] It also gave us a way to talk about achievement and hard work, about success and failure, and what it means to wake up the next morning and do it all again.
[4:40] And as a skiing family, we especially loved Eileen Gu.
[4:44] For those who don't know her, Eileen Gu is the most decorated free m-free-style skier in the Olympics history.
[4:53] She's also a model, and she's extremely smart.
[4:56] She just graduated Stanford while winning six Olympic medals on the way.
[5:02] Now, I saw she could ski, but once I heard her talk, I became a little bit of a fangirl.
[5:11] And most recently, Eileen was a guest on one of my favorite podcasts, Jay Shetty's on purpose.
[5:18] And she said something that stuck with me.
[5:22] Eileen said, hard work is super important, but overpraised.
[5:28] Direction underrated.
[5:31] She described somebody punching a wall.
[5:35] Working harder means punching the wall harder.
[5:39] Direction means noticing you could walk around the wall.
[5:43] But here's a catch.
[5:45] Punching harder many times feels like progress.
[5:51] And as a product leader, it made me think about what I'm seeing with AI right now.
[5:57] So many companies are using AI to punch the wall faster and harder.
[6:02] More code, more features, more experiments, and more demos than ever before.
[6:08] And the trap is there.
[6:10] Building fast feels like progress.
[6:13] Well, it isn't always.
[6:15] If I translate Eileen's analogy into product, hard work is building the product.
[6:22] Direction is deciding what we're building, for whom and why.
[6:27] So in product terms, it has never been easier to build something nobody wants.
[6:34] It used to be the opposite.
[6:52] But for us, for product teams, that cost was a forcing function.
[6:58] Because before asking a CTO or an engineering manager to invest, we really had to make sure that we are investing in the right thing.
[7:08] What problem are we solving?
[7:10] For whom?
[7:11] Why does it matter?
[7:12] And what are we choosing not to build?
[7:16] So we gathered evidence before building.
[7:18] Analytics, customer interview, market trends, mockups, business cases, of course, that didn't mean we always got it right.
[7:27] But the expense of building forced good questions much earlier in the process.
[7:37] Building is so much so much easier.
[7:41] An idea can become a working demo in a few hours or days.
[7:46] And honestly, I love that.
[7:48] Because building something and trying it out, it's my favorite way to learn something.
[7:54] And when I get a use to use a product or put it in front of users, I get signals I could never get from a deck or from a survey.
[8:04] But building only teaches you something if you're looking for something to learn.
[8:09] And sometimes we build a great demo.
[8:11] We love it, customers love it, everybody loves it.
[8:15] But in all that excitement, nobody asks the boring questions.
[8:20] Does this solve a real problem?
[8:23] Customers love it.
[8:25] Are they willing to pay for it?
[8:27] So here's my advice.
[8:29] You learned about all this building today.
[8:31] But when building, build to learn, not to commit to something.
[8:36] Ask yourself, what are we trying to learn?
[8:39] What's a hypothesis?
[8:41] What does success look like?
[8:43] Success isn't did we manage to build it?
[8:46] It's not even do customers love it.
[8:48] They can love 10 other products with the same thing.
[8:51] You have to decide what success is and then deliberately decide what deserves to be built.
[8:59] And honestly, building to learn isn't new.
[9:02] It has always been part of building great products, finding new ideas and putting them in front of users.
[9:10] In a way, I built my career around it.
[9:13] I was putting emerging technology into people's hands and then learning what happens and then building again.
[9:19] When I started my career here in Munich, I used to uh work at Hybris.
[9:24] I know there's a few people here that used to be there as well.
[9:28] And for those who don't know Hybris, Hybrids was acquired by SMP for about 1.5 billion dollars.
[9:36] And back then, experiments had a protected place in the company, a separate corner, hybrid slabs.
[9:45] We almost had carte blanche.
[9:47] We brought technology into physical shops using the same commerce APIs our customers render businesses on.
[9:56] One prototype I remember best, and I think you can relate was choosing wine, a digital wine shelf.
[10:03] Customers would have an application, they could choose their preferences, and then when entering a physical shop, the right wine bottles recommended for them would light up.
[10:12] Super neat, right?
[10:14] We loved working on it, it was super fun.
[10:17] Customers loved it, and retailers were really intrigued by it.
[10:21] But here's the thing a demo is something people admire.
[10:26] A product is something they get to depend on, and in many cases pay for it.
[10:30] And getting an experiment from the lab and the real business, that's where it got hard.
[10:37] I learned just how hard it was when I moved to Seattle and I started working in Xbox.
[10:47] I worked on game DVR.
[10:49] For those who are not gamers, uh gamers know it as Xbox Record That.
[10:54] So the idea is you game and then you do something really cool, you say Xbox Record That records less 30 seconds of gameplay, and then you have internet breaking rights forever because you can share it with your friends.
[11:07] At the same time, Twitch was taking off.
[11:11] Twitch is a similar idea but very different.
[11:13] You live stream your gameplay, you interact with others, why would anybody watch somebody else play?
[11:26] But millions of people did.
[11:28] And it was mostly a younger generation who grew up watching each other play.
[11:33] So for us, the hard part wasn't seeing the trend.
[11:34] Microsoft even tried to uh to acquire Twitch, that didn't work out, Amazon got them.
[11:41] But the hard part was the hard part was committing the business to it.
[11:46] Because for years, Xbox asked the question, what games should Xbox own exclusively?
[11:53] But now the question changed entirely, and it became where do gamers hang out?
[12:00] And that was a big shift and a really big question to ask.
[12:05] And a bet that size needed somebody at the very top to commit to it.
[12:10] And we didn't get that.
[12:11] So we did the default thing, which was continuing experimenting and building, and to be fair, we had some prototypes, but at that point, we never committed to do it until it was too late, but it's a different story.
[12:25] And that's the part that AI still doesn't fix today.
[12:29] You can have a team building Twitch overnight right now, but they're not gonna get the distribution and the network effects and turning the business around it and everything else.
[12:56] And Ned recognized the danger right away.
[13:00] He saw that once the acquisition was announced, developers were so worried about the news and Microsoft taking over the source code.
[13:09] And Ned said, you know what?
[13:11] We are going to make sure developers trust GitHub even more now.
[13:15] And he did it.
[13:16] Back then, in 2018, when I joined GitHub, there was 28 million developers on GitHub.
[13:22] Today, there's 228 million.
[13:26] That's like 8x growth from already a huge number.
[13:29] So obviously the developers still trusted us.
[13:32] And one of the really cool ways that we achieved that was by creating a little small organization, a small team called Special Projects, built to move fast and get developers excited about the future.
[13:46] I was on the team, and our job was win back developer trust.
[13:51] Ship the improvements the big core business couldn't prioritize.
[13:55] Try out risky and new ideas before the company was had to commit to them.
[14:01] That way, the company could bet uh could continue betting on strategic priorities and enterprise contracts, while we had an ever changing roadmap.
[14:11] We could pivot fast, and sometimes we had no road mob at no roadmap at all.
[14:16] And it worked.
[14:17] In one year, these are real change.
[14:19] We shipped over hundred changelog improvements with real feature requests that developers had in the past years.
[14:28] And we had betas that got tens of thousands of developers and enterprises interested within the first few days.
[14:36] But then there was a side effect.
[14:39] I couldn't see coming.
[14:41] We were stressed, stress testing the organization itself.
[14:46] Because when one team moved fast, every process built for a different speed starts to break.
[14:53] Release processes, documentation, lengthy reviews, social and comms team who had great people and good intention, but really long backlogs and needed days to build a copy for something we built in one afternoon.
[15:08] And here's the thing.
[15:18] Processes built for a different speed.
[15:21] But now AI gives every team that speed.
[15:25] And you can hear it, you heard it today so much in the room.
[15:29] And writing the code is no longer the bottleneck anymore.
[15:32] I heard about reviews, QA, different like industry-specific things.
[15:39] But one thing I want to focus today a little bit more is judgment.
[15:43] Some people like to call it taste.
[15:47] And when judgment cannot keep up with speed, something very predictable happens.
[15:54] We start building the wrong thing faster.
[15:57] Some call it AI slope.
[16:00] We come up with the wrong strategy, faster.
[16:03] Or the wrong feature, faster as well.
[16:06] Or we build and ship the right feature, but with a terrible UX that nobody nobody's using.
[16:13] And Atlassian surveyed this year about 1,000 product people, they do this every year.
[16:20] And 80% of them said AI is helping them ship faster, but customers don't see any more value.
[16:28] Now keep that 80% in mind.
[16:30] We're gonna come back to it a bit later, because there's another story.
[16:36] So far, I talked about the supply side.
[16:39] AI lets team build faster.
[16:43] But at GitHub, I watched AI change the demand side too, what customers wanted and how fast they wanted it.
[16:52] A few years ago, customers used to see a demo and ask, when will it be available?
[16:59] And we said, like maybe six to 12 months roadmap, and that was an acceptable answer.
[17:05] Not anymore.
[17:06] Now customers see a demo and they want to try it that day.
[17:11] And I'm not talking only about small customers, startups, tech unicorns that want to be on the bleeding edge.
[17:18] No, I'm talking about the most regulated customers we had, financial institutions, German automotive companies, healthcare, and so on.
[17:27] And why?
[17:28] That's because AI is promising returns that have never been seen before.
[17:33] And companies are so afraid of falling behind.
[17:37] So afraid that our customers didn't only ask how are we doing in terms of AI adoption, they asked, how are we doing against our competitors?
[17:47] Can we see industry benchmark of how everybody else is doing and show us where exactly in the industry benchmark we are?
[17:54] And the data shows it as well.
[17:56] Last year, IBM asked 2,000 CEOs in many countries around the world, and two-thirds admitted that they're so afraid of falling behind that they invest before even understanding the value.
[18:09] FOMO is real people, it is real.
[18:13] Now let's take an example.
[18:15] One that was a huge, huge learning to me.
[18:18] Two years ago, I was responsible for launching a GitHub Copilot Enterprise.
[18:24] At that time, that was our top AI offering at GitHub.
[18:28] And one of the favorite functionalities for us and also for the customers was AI-assisted code reviews.
[18:38] Now this is a bit of a table stake, I know, but it was two years ago.
[18:41] It was very new.
[18:43] And we put GitHub Copilot Enterprise in preview.
[18:47] We had a sign up wait list beta thing, and tens of thousands of customers, enterprises immediately joined the beta.
[18:55] So preview was not even something that was scaring people off.
[18:59] If anything, it was creating urgency.
[19:03] And back to this uh pull request feature to the AI assisted pull request.
[19:10] It was on our roadmap, but we couldn't quite figure out the right experience because models were not that great back then.
[19:17] It wasn't this December mod moment that they've noticed.
[19:21] They're okay, but not quite letting them review code.
[19:24] So while we were in meetings and Figsma and doing prototypes in VS Code and trying to figure it out and so on, we heard from one of our top customers, a financial institution, that they're already running a beta with a startup, a 10-person startup that was not even on the market, was already doing their pool reviews.
[19:44] And this was a huge learning for me.
[19:47] Because it meant that that was such a huge pain point for them that they were willing to risk and go with a startup.
[19:55] So two lessons.
[19:57] Lesson number one, build ahead of what the models can do today.
[20:02] Kevin Weil, who is a previous CPO at OpenAI, said that the model we're using today is the worst one you'll ever use.
[20:10] And that's really wild when there's a new model coming up every single week.
[20:14] Now the second lesson I learned was that startups have a big advantage.
[20:20] And that is they could make mistakes that we, a big company, couldn't afford to make.
[21:19] Those are the customers who depend the most on our products.
[21:24] And for me, that's been a central tension of building with AI.
[21:29] And building AI products too.
[21:32] For a product team, AI made building cheap.
[21:37] But promising didn't get any cheaper.
[21:41] GA still means security reviews, compliance, SLAs, support, docs, pricing, and the list goes on.
[21:49] And in many cases, especially in the enterprise space, a promise to keep it working for years.
[21:56] And the moment a customer depends on it, every shortcut we took to preview is our problem.
[22:03] And for me as a European living now and working in the US, this tension has been part of my career, and it shaped my personality in many ways.
[22:12] Because in the US, I learned how to move fast.
[22:15] And you know, Facebook did retire that slogan in 2014, side note.
[22:20] But I learned to move fast, take risks, do not take no for an answer.
[22:25] But growing up in Europe, I'm from Romania, and then I studied and worked here in Germany.
[22:30] I learned the craft of commitment.
[22:32] When you make a promise, keep it for years.
[22:35] And of course, these are cliches.
[22:37] There are plenty of American companies who keep their promises, and plenty of German teams, as we saw today that move really fast.
[22:44] But for years, many companies felt like they had to lean into one way or the other way of operating.
[22:50] And today that's no longer true.
[22:52] AI changed that.
[22:53] And to succeed, you need to do both.
[22:57] Move fast and then decide what you want to commit to.
[23:01] So let's zoom out.
[23:03] In this crazy world, everyone can build.
[23:07] Customers won't wait.
[23:09] Dave asked us this morning when everyone can build what makes a great product.
[23:15] And I want to take it one level further.
[23:18] When you have a great product, where does product advantage even hold?
[23:24] Let's look at three places.
[23:26] How adoption happens, what value really means, and what actually lasts.
[23:35] For years, I've been fascinated how software adoption happens, even in companies.
[23:42] Companies sign the contracts, but in many ways, people decide what gets used on a daily basis.
[23:49] And I saw this app close at GitHub.
[23:52] When GitHub was taking off, it spread bottom-up.
[23:55] You probably, many of you are in that crowd.
[23:58] Developers loved it, they used GitHub in their side projects on the weekend in the evenings.
[24:03] And then they couldn't imagine living without it.
[24:06] So they brought GitHub to work.
[24:08] And then little teams in companies started using GitHub before anybody officially approved it.
[24:14] A bit later, GitHub understood what's happening and they offered the enterprise plan.
[24:19] And I wasn't there back then.
[24:21] And the enterprise plan included governance, security, license allocation, and so on.
[24:27] So at that point, it was a no-brainer for companies to sign the contract and make it a standard across the organization and then push it top down.
[24:36] Now, in the last two or three years, we saw the same thing happen again with tools like cursor, cloud code, and codecs.
[24:46] People tried it at home, probably many of you did, and we loved it.
[24:50] And once we had a glimpse of how the future could look like, we couldn't imagine going back to work without those tools.
[24:57] So people brought these tools at work again.
[25:01] And the data backs it up.
[25:04] Today, more than half of US companies are buying AI tools, and this data is a bit outdated, so it's way more today.
[25:13] And yet, two thirds of AI use happens of personal accounts at work based on like what people want to use.
[25:23] So in the US, I watched compliance teams changing their job.
[25:28] Instead of telling everybody what tool to use and what they cannot use, now they're giving access to all the tools safely.
[25:35] And last year I was here at BMW Welt talking to a lot of our German customers at GitHub, and I saw the same shift happen.
[25:43] Even here, there are so many companies trying out all these tools, and the compliance team's job was making sure that they can use the tools within the guardrails.
[25:52] So that's the same two instincts of moving early and learning while keeping promises and being really safe.
[26:01] So let's go a bit back.
[26:03] Remember the 80%, the one with teams not seeing value any sooner.
[26:08] Let's talk about what value means.
[26:11] Even after a company signs a large deal or adopts a product, it's really easy to call it success.
[26:18] Our job here is done.
[26:20] But that's not right.
[26:22] Because sometimes we mistake activity for adoption.
[26:25] And when what we really care about is impact and outcomes.
[26:30] I learned this at GitHub once again, step by step.
[26:33] So back in the day, GitHub Copilot was one of the first AI tools to reach mass adoption.
[26:39] People don't know it, but it came before ChatGPT became famous.
[26:43] And it spread like wildfire.
[26:45] It was kind of wild.
[26:46] And it was not only because the product was really good, it was.
[26:49] There was also first to market.
[26:52] But it was so easy to use back then.
[26:54] It was an AI autocomplete on steroids.
[26:57] You just press tap and the code was getting written.
[26:59] Super easy and in people's workflow.
[27:02] As models got better, chat was the next evolution.
[27:06] And you probably remembered it.
[27:08] Every company had like a little chat bubble on the right, and then they called themselves AI integrated or AI powered.
[27:17] But chat was a little bit uh slower to adopt.
[27:20] And that's not only because it wasn't a good product, it was actually quite smart and quite good.
[27:26] But it changed to ask people the it asked people to change how they work.
[27:32] They had to stop, switch context, copy paste, start a conversation.
[27:36] And the payoff wasn't always obvious at that point at that point.
[27:41] And up until that point, all the conversations were about adoption.
[27:46] What's our adoption numbers?
[27:48] Can we get it to 70, 80%?
[27:50] How do we compare it to others and so on?
[27:52] But then models got better.
[27:55] And people started delegating little tasks and whole tasks to agents.
[28:00] And the price got higher too.
[28:03] And boards and CFOs started asking different questions.
[28:07] Instead of license activation or company wide adoption, they started asking about impact.
[28:13] What's the return on investment for all this money we are spending for AI tools?
[28:20] And that question changed the way I think about success metrics.
[28:25] And it goes back to first principles and one of my favorite questions as a PM.
[28:30] What is the problem we are trying to solve?
[28:33] Because if we really know the problem, then we can design success metrics that truly measure the impact and the progress against that problem.
[28:41] And that's usually an impact metric, the same metric we can share later with a customer or with a board.
[28:48] So going back to the same example, the AI code review feature I loved.
[28:53] When we launched it, it was very clear what's the problem statement.
[28:58] Pull requests are getting stuck in the merge in the review queue, and they stay there for hours and sometimes days.
[29:05] And we knew that with AI coding and assistance, that's gonna get even worse.
[29:10] And in fact, yesterday evening I saw on X a stat that pull request numbers grew 3x in the last year, and a third of those are done by AI agents.
[29:22] So it kind of checks out.
[29:24] We had this problem at GitHub, customers had it too.
[29:28] So at launch, instead of pitching a feature, we had clear data proving the impact of the feature internally at GitHub and with a few pilot customers.
[29:38] Pull requests were indeed going down, uh not uh going back to taking a few days, a few hours to reviews and to merge, and no longer days.
[29:47] And that way of pitching things really resonated to our customers.
[29:52] And it became a little bit of a best practice every time we launch something new, show the feature and the outcome of the feature.
[30:02] And this shift from access to usage to impact, we see it in pricing too.
[30:09] Now we're well into the per seat pricing model.
[30:14] That's access.
[30:16] We were there.
[30:17] We moved to per token, and I think that's what we many of us experience right now when we run out of the tokens that we get from our seed allocation in cloud, we have to buy extra ones, these are token pricing.
[30:28] But some companies, you know which one, Intercom, is now charging per outcome 99 cents per ticket solved.
[30:39] You only pay for a job well done.
[30:42] But if outcome is what ultimately matters, you only have space for those features that drive the outcome.
[30:49] Everything else doesn't matter.
[30:52] And today we saw you can add a feature almost for free.
[30:55] Everybody can.
[30:57] But the catch is our heads haven't gotten any bigger.
[31:01] If anything, we're constantly distracted and have the memory of a goldfish.
[31:07] Users can only hold that much in their head, and so can the leaders deciding what product they're going to buy.
[31:14] And Naniu, who works at Codex, uh, they used to be before at Linear, said something I loved.
[31:21] The biggest constraint is people's ability to understand and absorb what you're giving to them.
[31:29] AI is amazing at generating possibilities, but it's so bad at telling you you have a bad idea, you probably noticed that as well.
[31:37] So one of the most important skills is deciding what to take away, what to kill, what to combine, and what to simplify.
[31:47] And one should never even make it into the product to start with.
[31:52] Now, Trader Joe is one of my favorite examples.
[31:57] For those who haven't been, Trader Show is a well-loved grocery chain in the US.
[32:02] They have small store small stores, quirky own brand products, and a little bit of a cult following, including me.
[32:11] And fun fact, they're actually owned by the Aldi Nord group, so there's a little bit of a German DNA there as well.
[32:18] A typical supermarket in the US has 50,000 products.
[32:23] That's wild.
[32:24] Trader Joe has 4,000.
[32:28] But check this out, Trader Show sells roughly four times as much as a regular supermarket per square meters.
[32:37] People trust TJs will only show them the best options.
[32:41] And by the way, these are really good.
[32:43] So if you make it to a TJ, you should buy them.
[33:15] A few weeks ago in Seattle, um Charles LaMana, who runs Copilot and Agent at Microsoft, put it this way.
[33:24] If my agent interacts with your app exclusively, the agent is your customer, not the end users.
[33:31] And agents are really harsh customers.
[33:35] His view is that most software in business, but if you ask me, even in the consumer business, will run headless behind an assistant.
[33:44] And I heard from many people here already that they use Instinct for doing doctor appointments.
[33:53] Agents are deciding which tools to use for a given task.
[33:58] They don't see your onboarding, which might be great, or your design or your beautiful landing page.
[34:03] They see whether the tool call worked and return the expected results.
[34:09] So, in a world with so many choices, where every week there's new products, new AI models, and new personal assistance.
[34:19] What makes an agent or a person pick our product and keep picking it?
[34:27] Honestly, it's too early to tell.
[34:30] The best practices have not been written yet.
[34:33] I believe it's no longer about being first to market.
[34:37] Everybody can now reverse engineering anything.
[34:40] I believe the advantage goes to the people who learn fastest.
[34:45] And learning fast means pushing the limits, building where things might break, and trading every failure as data, and then turning all those learnings and data into things that compound.
[34:59] We talked a lot about context today, what your customers taught you, all the data that is not accessible by an LM.
[35:06] Trust, a dependable product, a dependable team, the CISO, who said yes, and the trust that you built for years.
[35:15] And then community and network effects.
[35:17] The people, their photos, their stories, their code.
[35:22] You can copy features, but we learned you cannot copy where people are.
[35:27] And here's what I learned the hard way.
[35:30] Even if you have all those things, context, trust, community, none of it matters if you stop pushing the limits, or if somebody learns faster than you.
[36:04] What is the problem you're trying to solve?
[36:07] But this year I realized my answer evolved a bit.
[36:12] Because AI made building so cheap.
[36:15] So my advice is to build.
[36:17] Push it until it breaks.
[36:19] See where are the limits, because that's where the learning is.
[36:22] Then it's your judgment, your customers, experience, data, your teams to deserve what deserves a promise.
[36:32] And for me, that's not just advice, that's where I'm right now.
[36:36] Because this summer I left GitHub and started a new chapter.
[36:39] I'm building again hands-on.
[36:41] And I'm doing exactly what I told you.
[36:44] I'm building to learn, pushing until they break.
[36:48] And although I wrote this talk myself, the trap is still there for me.
[36:53] Some days I'm building something and I get so excited.
[36:57] And then I have to remind myself wait, just because this works, this is not progress.
[37:02] This is not learning.
[37:07] And I'm learning a ton along the way.
[37:10] Which brings me back to Eileen.
[37:12] The hard work and the building has never been easier, but the direction going around the wall is still on me.
[37:20] And every week on Friday, I'm not using Dex, but I'm using I have my own chief of self.
[37:26] I'm gonna move to Dex soon.
[37:28] Um I use a I do a retro with myself or with Claude.
[37:31] And I ask myself, why did I build this thing that pushes the limits?
[37:36] And what did I learn?
[37:38] So that's my question for you for next week.
[37:41] What is the one thing that pushes the limits with a goal of learning fast?
[37:46] Because in the end, it comes down to one two things.
[37:51] Move early enough to learn, and then commit carefully enough to endure.
[37:55] Thank you.
[37:56] Thank you so much, Simina.
[38:06] Oh, I love the stories.
[38:03] I I simply loved it.
[38:10] Um I think.
[38:13] Do we have time for a question?
[38:15] Simina, you will be in the bar later, right?
[38:19] Find Semina in the bar.
[38:21] I will remind you again.
[38:23] Drinks are on us tonight, not cocktails.
[38:26] If we fill this up, if we fill this up next year, so that we have about like 30, 40 seats left.
[38:33] Um if we fill this up next year completely, then also cocktails are on us.
[38:38] So I want you.
[38:41] I want you to keep this in mind.
[38:44] Sorry, sorry, wait, wait, wait.
[38:45] Can you come back up again?
[38:46] Sorry, sorry, sorry.
[38:47] Before I keep rambling.
[38:49] Huge, huge roaring last applause for Simina Passad.
[38:54] Thank you so much.
[38:57] You are not sneaking away from me without an applause and the present.
[39:01] Alright.