# When Everyone Can Build, What Makes Great Product?

Dave Killeen, Just Product 2026, 14:45. Automatic transcript of the livestream (Parakeet ASR, unedited, expect misheard names). Timestamps are positions in the talk video.

[0:01] So the next speaker.
[0:08] It's also the first in our theme AI transformation and leadership.
[0:14] And let me tell you a bit about Dave.
[0:17] So I think Dave and me, we met the first time last year at Pandemonium, I think.
[0:23] And I saw his talk and I thought, oh wow, I really want to get that energy on the stage at just product.
[0:32] And then he did another thing, um, beginning of the year.
[0:35] He basically built an AI chief of staff called Dex.
[0:39] And I started using it uh back in February, and it has completely changed how I worked, honestly.
[0:46] Right.
[0:46] So telling you about one moment, like I was using it like two or three weeks in, and I would was doing all my tasks in it, like daily planning.
[0:54] And then on a Thursday, I can still remember that on a Thursday when I was doing my daily plan.
[0:59] The agent said, Hey Sebastian, you're giving a talk next week, uh Thursday.
[1:03] Do you do you think you might want to start preparing it?
[1:07] And it was like worded in that way as well.
[1:09] So I haven't seen you doing anything for this.
[1:11] So do you do you want to maybe start?
[1:13] And I'm like, oh yeah, that's actually a really good point.
[1:16] Yeah.
[1:16] So this has seriously changed how I work now.
[1:19] So and then we keep writing, going back and forth, and say, Yeah, do you want to come to Just Product?
[1:24] He said, Yeah, yeah, yeah, I want to do it.
[1:26] So he's here now.
[1:27] I'm super happy.
[1:28] Please giving a huge applause.
[1:30] Dave Killeen.
[1:37] Thank you, thank you.
[1:38] Hello, hello, hello everybody.
[1:40] I hope you're all keeping well.
[1:41] Good gitty god.
[1:42] This is what a screen.
[1:44] It's way bigger up here.
[1:45] I feel like a dwarfed Irish leprechaun.
[1:48] Now, look, some of you are probably tired, but some of you also in the front, eagle-eyed might be going, good giddy God, Dave is up here wearing smart glasses.
[1:58] He's either gone full tech pro cyborg, or he's recording us to train his AI.
[2:05] I'm actually doing both.
[2:06] Yes, I'm beaming my speaker notes into my eyeballs.
[2:09] Why?
[2:10] We've only got 35 minutes, 30 minutes I should say, and I'm an Irish geek and I love a tangent.
[2:16] So honestly, this is all for your protection.
[2:19] I want to talk about um two connected journeys today.
[2:23] First, I want to talk about, let me get this done first.
[2:28] So I basically want to talk about how I'm compressing and this side project, um, then the product development life cycle, I should say, and uh as an open source side project that I build my spare time.
[2:40] And second, what happens when you try and bring those same ideas into an established company like Pendo?
[2:48] The question underneath all of this, though, is a bigger question for everyone in this room.
[2:53] What happens to when tiny teams can suddenly build really good software incredibly quickly on a fraction of our cost space that many of us here are carrying?
[3:03] What does that do to pricing?
[3:04] What does it do to packaging?
[3:06] And the things that we used to think were defensible, right?
[3:10] And if you already have what new entrants like me desperately want, which is customers, distribution, trust, and a position in the market, how do you remove the organization organizational drag around your advantage and start moving just as quickly?
[3:25] And so, as Field CPO at Pendo, I spend my time with executives at companies from SAP to MasterCard, through to some of Europe's fastest growing startups, and we talk about product craft, AI fluency and transformation, which means I also get to play with AI at the edge of what's possible, hence these damn glasses.
[3:43] And I'm unhealthy obsessed with it, much to my husband's fucking frustration, frustration.
[3:51] But before we start all of this, please don't feel that you're behind.
[3:54] Because after two years, two years of bottomless AI budgets, almost everyone I meet is asking the same damn question.
[4:01] Where is the bloody return?
[4:03] And how do we actually change the way the company works?
[4:06] So, okidoki, let's do this.
[4:09] I am not going to hand you the comfort blanket that everyone seems to be wrapping themselves around right now.
[4:17] People say, sure, AI can write the code, but it will never, it'll never replace human taste.
[4:23] There's an inkling of truth in it, but I think it's a massive get out of jail card.
[4:28] And I should know, I was one of the loudest people selling it.
[4:32] I used to go around advocating that we were all the Michelin starred head chefs, right?
[4:39] AI is our sous chef doing the heavy lifting.
[4:42] But we designed the menu.
[4:44] We drizzle the jus.
[4:45] We place that piece of kale on top, make the plate pop, and I was utterly convinced that taste was our moat.
[4:52] And I feel I was wrong.
[4:54] In December 2025, AI hit a turning point.
[4:58] Andrei Kirpathy, the godfather of vibe coding, said that it had finally passed the threshold of coherence.
[5:04] What a geeky thing to say, right?
[5:06] He meant the barrier to entry wasn't code anymore, it was upfront clarity.
[5:10] And around the same time, Claude Code introduced what's called session start hooks, and others quickly followed.
[5:16] And what that meant was instead of every chat starting like a clueless numpty, you could load useful context into the start of the next channel.
[5:26] So developers then started using it to feed in a lessons file, a markdown file from the last session.
[5:31] Very simply, a file where the AI puts its hands up and says, Yes, I'm so sorry, Dave, I made all those mistakes, I'm a bloody idiot, I won't do it again.
[5:39] And so then what you do is you load that into the next session, and then hopefully cross your fingers it doesn't do it again.
[5:44] And they called that compound engineering.
[5:48] These two things together changed the pace dramatically.
[5:52] The models got more capable, and the context around them started to compound rather than resetting every time.
[5:58] And that got me thinking, why should that apply to software engineering?
[6:02] What if every conversation with my AI started already knowing what my goals are, what my projects are, what the people I work with?
[6:09] What if actually from that context it got more useful week after week every time I danced with it?
[6:14] So compound engineering, but for knowledge work.
[6:18] And that's what led me down the following rabbit hole.
[6:21] I built myself DEX.
[6:23] DEX is a personal, um, deeply personal AI operating system stored locally on my MacBook that pulls together my goals, my projects, my people I work with, all as markdown files, and helps me work out what the hell to do every day.
[6:37] Every morning it looks at what I said mattered and it keeps me accountable.
[6:41] I even gave it my colour profile.
[6:43] I say I'm very yellow, right?
[6:45] Ideas, ideas, ideas.
[6:47] And I have a bit of OCD, so I plan my week in a Friday.
[6:50] Monday morning, I bounce into Dex like a bloody JRSL bunny, fizzing with ideas.
[6:55] And basically I get the following, Dave, we love that you're yellow.
[6:59] We love it.
[7:00] But bloody hell, Dave, you're already off on three tangents.
[7:02] It's 10 past 10 on the Monday morning.
[7:04] Come the hell down, Dave.
[7:05] Don't worry, Dave, I banked your ideas, Dave.
[7:07] Now, if my husband talked to me like that, I would tell him where to go.
[7:11] When Dex tells me, I'm like, thank you, Dex.
[7:14] I love you, Dex, right?
[7:16] So it was stunningly useful.
[7:19] So much so, I open sourced it.
[7:22] I thought maybe, maybe, maybe some other people on GitHub might find it useful.
[7:26] Uh now I'm not technical, I do not build software, at least I didn't think I did, but three weeks later, I got this on LinkedIn.
[7:32] I was in the toilet seat, I kid you not.
[7:34] The head of AI at one of the world's largest tech companies, pretty much top five.
[7:39] I'll count, I'd love to tell you who, but I can't, got in touch, and as you can read behind me, they wanted to know if their executives, yeah, not everyone, just execs, could use it instead of a tool, they had built with 50 engineers over the course of a year.
[7:53] Now, nothing sadly has come of that yet, otherwise it did, I wouldn't be here, quite frankly.
[7:57] But that kind of interest puts an absolute rocket up your R.
[8:01] And building fast taught me something that I hadn't quite expected.
[8:05] When you have AI in the kitchen, you do not just get code faster, you get an overwhelming explosion of utterly bonkersly brilliant ideas.
[8:16] But when hundreds of ideas hit you at once, your taste becomes a bloody bottleneck.
[8:21] So being a head chef isn't enough anymore.
[8:25] You've got to go outside, you've got to become the truffle hunter.
[8:29] The head chef judges what's on the plate, but the trouble hunter decides what's worth cooking in the first place.
[8:34] Same taste applied a mile further upstream.
[8:36] So our job becomes going out into the muddy forest in my case, pattern matching at massive scale, watching what the nerd community is building and being ruthless about which golden nuggets do I pull out of the dirt.
[8:48] But sure, who in this room has time for all that mad malarkey?
[8:52] We simply bloody don't.
[8:54] So I send out what I call my dog robots.
[8:57] I built a system that hunts for me.
[9:00] So I say in plain English, hey Claude, hi, Dave here again, every morning, please have a mooch on X.
[9:05] Please tell me what everyone's rabbiting on about and find some new open source nuggets on GitHub that's relevant for me that I can steal.
[9:12] Thank you, Claude.
[9:12] I love you, Claude, I love you.
[9:14] Always say I love you.
[9:15] One person with the open source software ecosystem scanned every single day.
[9:20] That is not a productivity hack.
[9:23] It's a new operating model because when intelligence is effectively free, the advantage no longer becomes who has the most data, the most engineers, the biggest roadmap.
[9:30] The advantage becomes the speed of the loop.
[9:34] So, how fast can you spot the signal?
[9:37] How fast can you turn that into software?
[9:39] And how fast can you prove that customers actually give a damn?
[9:43] Small teams now, like me and others, are already making that loop lightning quick at a fraction of our collective cost base, right?
[9:50] So I'm not here though to be a Debbie Downer.
[9:54] Once you see the shift clearly, the opportunity, as many have been saying here before me on stage, it is staggering.
[9:59] We just need to change how we play.
[10:01] So for me, it's three moves.
[10:03] Number one, out hunt.
[10:05] Number two, bend the system, and number three, prove impact.
[10:10] Out hunt is how much can we spot the signal before it's obvious?
[10:14] Bend the system, how can we reshape the work or it gets done around AI and prove impact, obviously, you know, is when building gets cheap, the question stops becoming can we build it?
[10:24] It becomes should this even bloody well exist.
[10:26] So let's start with out hunt.
[10:28] Hello, GitHub.
[10:30] Most of us think that this is where engineers, where the geeks keep their code.
[10:35] I want you all, please, please, please, to see it as an early radar on your market going forward.
[10:42] Before something becomes a product category or hits an analyst report, it often starts as a here as a scrappy repository.
[10:49] Truffle hunting isn't reading more newsletters, it's watching where the market is actually being built.
[10:54] GitHub shows you where the truffles, quite frankly, are growing.
[10:59] There's a page on GitHub called GitHub Trending.
[11:02] It's a real-time read what the engineers are building right now.
[11:05] So it's tomorrow's competitive landscape, today.
[11:08] But it only shows you a small selection, about 20.
[11:11] And I want everything relevant to DEX and the people I'm building it for.
[11:14] So there's no bloody way I'm going to do that by hand.
[11:18] So I have an agent every night, it knows my ICP, my ideal customer profile, it knows my vision for DEX, it knows what we've built so far, it knows what's on the backlog.
[11:26] And while I sleep, it scans GitHub to find things relevant to that context, right?
[11:32] We talked about it.
[11:33] It gives me a brief.
[11:35] What matters, why it matters, what might be worth adding to in the backlog.
[11:39] Yes, I have a shared ideas backlog with my bloody AI.
[11:45] Honest to God, folks, we compete for the best ideas on different vectors.
[11:50] When people ask me, is AI creative?
[11:47] I think it's the wrong question.
[11:54] Am I more creative because of AI?
[11:57] Absolutely one million percent.
[11:59] One seed from the agent that you see behind me can spark dozens of ideas within me.
[12:03] And let me tell you, when the AI competes with you on ideas, it does something to your head.
[12:08] Historically, the bottom 80% of a roadmap is crap, as mentioned earlier.
[12:13] That's why we never get around to it.
[12:15] But when your backlog turns into 200 genuinely ideas, it becomes overwhelming.
[12:20] 200 brilliant ideas is not a roadmap, it's a coordination problem, it's chaos.
[12:25] It's a system of work problem that I'll come back to in move two.
[12:29] But let's keep nerding out more on open source.
[12:32] I don't think we pay much as attention to it as we should do as product managers, if I'm honest.
[12:37] So my dog robots were out bringing back feature ideas, and not just feature ideas, they sometimes they bring back an entire damn product.
[12:45] On the left, what you see here is a fully functioning open source CRM system called 20.
[12:50] It is stunning.
[12:51] And on the right is an open source alternative to clawed design built by one developer, a few thousand dollars of credits back in the day, probably nothing now, and starting line has moved.
[13:01] He built that in one week, by the way.
[13:03] Features that used to feel expensive and defensible are becoming building blocks for other people to repurpose.
[13:09] It is utter bloody madness.
[13:11] So the stands keep shifting.
[13:12] Look at Jev, which came out the other week.
[13:15] Yeah, went viral in September.
[13:17] Its co founder, who's the co-founder of Chat GPT, he says the team spent two years building it in stealth.
[13:22] Two years.
[13:23] And within days, open source alternatives are appearing and getting quite close to performance.
[13:28] And then when the very same Grockbok came out from SpaceX AI, you've got that behind me.
[13:32] That's that also came out a week later.
[13:34] The time between what's novel and there are alternatives is collapsing, which is brilliant if you're a builder, and pretty unsettling if you're guarding an established position.
[13:44] And I'll be brutally honest, I feel it too with Dex.
[13:47] Some weeks I look at the damn thing and I'm like, oh my god, like I'm nearly 50, I'm wasting my life.
[13:52] This is ridiculous.
[13:53] What are you doing, Dave?
[13:54] Honestly.
[13:55] Which points to the real problem with move one.
[13:57] The problem becomes the actual concept of what a moat is.
[14:02] There was a recent uh line from George Stevenson at Index Ventures questioning whether a moat is still intellectually honest.
[14:10] If we imagine it as something that you dig once and then you sit behind forever.
[14:14] And I'm not saying moats are dead, right?
[14:16] Quite the opposite.
[14:17] Distribution matters, customer relationships matter, trust, propriety, data, network effects, brand.
[14:22] All of this matters.
[14:23] In fact, many of you here today have something I don't have with DEX.
[14:26] You've got distribution.
[14:28] You already have customers who know you, a brand they trust, a place in their working day, a market.
[14:32] Position and you engine would just be dying to have.
[14:35] So the threat isn't the startups win, everyone else go home.
[14:38] The opportunity is what happens if you all combine those existing advantages with some of the ways much more AI native teams are beginning to work, where they automate more of the hunt for the signal, where they codify what works, where they run more execution in parallel that we saw here today, when they get the idea from customer evidence to out the door to dramatically faster.
[15:01] So the real question is how do we remove the organizational drag, the friction around the advantages that we in this room already have, advantages that new entrants are desperate for.
[15:13] And so that brings us to move two.
[15:16] If the idea and the build aren't enough, what is the edge?
[15:21] I think it's about building the system of work.
[15:23] And it's this is the heart of it for me, absolute heart of it.
[15:26] If one part of your organization suddenly becomes dramatically faster, the work doesn't magically disappear, the bottleneck moves.
[15:34] Engineering gets faster, perhaps product can't decide quickly enough.
[15:37] Product gets faster, perhaps legal become the cue.
[15:40] Or you generate far more ideas, but bloody hell no one has enough evidence to know which deserves investment.
[15:45] So bend the system means continually asking, continually asking.
[15:50] Where is the work getting stuck now?
[15:51] And how do we redesign the entire system of work around that new constraint?
[15:56] Because there's a useful analogy in this, by the way, and bear with me.
[16:00] We need to talk about shipping containers.
[16:02] There's a bloody brilliant book called Reshuffle by Sonji Paul Chowdery.
[16:06] He argues that we misunderstand AI when we treat it as simply as a productivity tool, a faster horse.
[16:11] He compares it to the shipping container, and the container didn't just make shipping cheaper, it reconfigured global trade.
[16:17] The value wasn't just the box, it was the reconfiguration around the box itself.
[16:22] And that's the mindset shift with AI.
[16:25] Redesign the system, not the individual task.
[16:28] AI represents a complete redefinition of how work gets coordinated across an organization, how we win, how we compete.
[16:35] But unfortunately, most organizations, from what I can see, aren't setting the teams up for success.
[16:41] At one extreme, you buy licenses, you hand people a blank chat box, you cross your fingers, you hope.
[16:45] What should be doing instead is redesigning the environment that people work in.
[16:50] And later on, I'll show you how Pendo is doing exactly that, redefining how its work gets done, childry style across 500 people within the PDLC.
[16:58] But first, my own journey.
[17:00] And the reason for doing this is not to go all Dave heavy and overwhelm and drive up anxiety.
[17:05] It's just to show you how I genuinely very strongly feel small teams will increasingly work far, far faster on a far lower cost footprint, and that has huge implications for our pricing and packaging and how competitive the rest of us can stay when we have larger organizations with more mouths to feed.
[17:22] So my dog robots and I were bringing back more truffles than one could ever cook.
[17:28] I needed software that could coordinate multiple agents in parallel efficiently, without me losing the will to live from having to mother each and every one of them.
[17:39] Agent orchestration was largely an unsolved problem.
[17:42] I tried Linear, I tried JIRA, I tried Symfony from OpenAI, and back in May, June, there was nothing else really.
[17:49] And so all of these tools back then were built for human teams moving at human speed.
[17:53] And I needed a kitchen that would work with my AI and their kitchen staff, and it just didn't exist.
[17:59] So I had a chat with Claude, hey Dave here, back again, sorry, and I ran to develop my problem and I built one, which is exactly the whole point of malleable software, right?
[18:08] When the tools don't fit the way you work, you stop waiting for someone else's roadmap and you bend the system.
[18:14] You build your own software factory.
[18:16] And this behind me is what I showed to a thousand product managers in London at mine, the product in June.
[18:17] On the left was the workflow.
[18:24] Every column invoked a different skill that you see behind me.
[18:26] So the first column that you see is the raw ideas column.
[18:30] So that's whatever AI and I dreamt up.
[18:33] And then the 10x column, the next one, is what I also call the Steve Jobs column, which is a skill I created where everyone else's ideas, shit, including mine, just please 10x the shit out of this, right?
[18:44] And then the third column is the agent ready column that's next to it, which takes the stronger ideas and turns them into work that an agent can actually execute with a clear way of proving very important to prove whether it's finished, right?
[18:57] Most important thing ever when you're working with AI, you know, verification.
[19:00] So you move a card into a column, and then the relevant clawed code skill gets activated in the back end.
[19:05] And that on the right was my little virtual product team.
[19:09] I defined these agents much as I would define roles in a real team, product, design, engineering.
[19:14] Each of them had a clear remit, a point of view, an understanding of what could look like.
[19:18] And I wasn't just throwing the same generic AI at every problem, the roles gave a purpose, a helpful skew that was better than the average.
[19:27] And so I had a workflow.
[19:29] I had specialist agents, I had a system for moving work between them, but the real magic wasn't the board, and it wasn't even the agents and the skills, it was what they all knew.
[19:38] It was the context that they all shared.
[19:41] Behind me are what I call my truths, the things that I currently believe will be true about the world in 12 months from now.
[19:47] And I increasingly think every leadership team, every product team, every company, quite frankly, needs some version of this.
[19:54] The assumptions up the stack that Martin brilliantly called out earlier, right?
[19:59] Get his book, by the way.
[20:00] It's phenomenal.
[20:01] What do we believe will change about our customers?
[20:05] Our market, the technology, the economics of business.
[20:08] These are not just predictions, carved in stone, they're actually hypotheses which we argue with and we update as the evidence changes, right?
[20:16] And because the roadmap tells you what you're doing, you're true to tell you what future you're doing it for.
[20:20] And so I thought to myself when I show this back in June at Minded Product London, well done, Dave.
[20:25] Bingo, this is amazing.
[20:27] You've nailed it, you can rest.
[20:30] But here's a slightly embarrassing bit.
[20:31] That whole thing I've just shown you in those last two slides, it's all in the bin.
[20:35] Because back then, I was still the truffle holder, traffic controller.
[20:39] I moved an idea from one column to the next.
[20:42] I invoked the right skill, I handled the work to the next agent, and the models got dramatically better at working for far longer and coordinating with each other.
[20:51] So I became that bottleneck.
[20:53] The valuable bit wasn't the functionality of the board that you see, it was actually the context that changed that that was in within it.
[21:01] And the orchestration layer is what I wanted to focus on next is how do you get that work coordinated?
[21:06] And I so I created my own shipping container.
[21:09] The old way of working quickly had me as the bottleneck.
[21:13] And so today I use this tool called herder.dev.
[21:16] And it's an open source orchestration skill, also with it called First Mate from an ex meta-level eight engineer, really good on GitHub, and it sits above it.
[21:25] And don't worry about the names, by the way, but the shape of the system is what you should look at.
[21:30] I stay in one plain, calm chat, and basically have a nice calm conversation.
[21:36] And then multiple agents, they all just matter to each other, have no idea what they're doing.
[21:40] And the system brings me back the decisions that me, I need to make, right?
[21:44] The judgment, the hunting, making bets, I've got more time then for that, right?
[21:47] And this is what I feel we'd be increasingly up against.
[21:51] One person or a tiny team with an AI crew building increasingly in parallel on a cost base of a few hundred or a thousand dollars or so a month.
[21:59] Their ability, therefore, to undercut uh larger products on pricing will become increasingly problematic for the industry.
[22:06] So that's me, that's one person, evenings and weekends, huge highs, huge lows, borderline divorceable.
[22:13] And now for the harder question.
[22:16] So, Dave, how do you make this work across 500 people at Pendo?
[22:20] Our engineers, they leaned heavily into AI, and throughput jumped up 50% to start with, which is fantastic.
[22:27] But the bottleneck then moved to product and design.
[22:29] We had to basically decide fast enough to keep feeding the engineering machine.
[22:34] Teams built their own AI skills and their own workflows individually, some were bloody brilliant, but they were all different everywhere.
[22:40] And so the leadership team are reviewing 30 squads, they had to play this chameleon game just to decode each team's language.
[22:47] We needed to turn individual brilliance into company-wide capability, which brings us to meet Rex.
[22:55] Pendo's shared AI chassis.
[22:58] Our mascot at Pendo has always been a dinosaur, so now my solo side project X finally has a big brother.
[23:04] Rex grew out of a simple realization that AI fluency is a sharing problem, not a training problem.
[23:11] Rex is a shared catalogue of reusable skills, ways of working with AI.
[23:15] We call them skills and they sit behind an MCP server that we host internally.
[23:19] So when someone is working in Claude, Rex brings the relevant skill into the job from the MCP server.
[23:24] Yeah.
[23:25] We standardize how the company works, not which AI model people use, they can use whatever, and it's not a static prompt library either.
[23:32] If someone finds a better way of improving it, they can recommend it, and it gets published by the AI.
[23:31] And so everyone's baseline improves together.
[23:39] Let me show you.
[23:39] One of the skills we have in Rex is called the conviction document skill, where we're not asking AI for prettier PRD, it pulls together the evidence around a bet.
[23:49] Customer insight, product data from Pendo and market context.
[23:53] What problem are we solving?
[23:54] For whom?
[23:55] What would change our minds?
[23:56] This isn't a PRD, it's a wager.
[23:58] And then comes the part behind me, one document and many outputs that feed into it, right?
[24:04] The same conviction document feeds the product spec, the engineering brief, the OKRs, the board slide, the GTM narrative, the sales enablement.
[24:12] We establish the conviction once and we let the context travel with the work.
[24:17] Which is beautifully tidy on a slide like this, but you still have to get people to actually work that way, right?
[24:24] So how do you do that?
[24:25] Well, we are forcing the issue with Headless October.
[24:30] Hence our headless friend behind me.
[24:32] Nobody at Pendo is allowed to log in to Pendo for the month of October.
[24:38] Not a single person.
[24:39] If they do, we'll slap their wrists.
[24:41] And they'll do their dance instead in Claude through Pendo MCP, Rex, and other MCP connectors.
[24:46] And full disclosure, we have not got an idea, not a foggy how this will go.
[24:49] It might be a car crash, which is more than fine because we're willing to break some eggs.
[24:54] My challenge to you all here is find your forcing function for the change you know in your hearts, in your bones, that you need to make.
[25:00] But even if you get all of this right, one massive question still hangs.
[25:04] And that question isn't can we build it?
[25:06] Is can we prove it?
[25:08] Can we prove any of this is actually creating any value?
[25:09] And that's move three.
[25:10] And this is where I have to admit something slightly embarrassing about DEX.
[25:14] I haven't got a foggy, not a bean, not a clue who's using it.
[25:18] Honestly.
[25:19] Out there since January, DEX works in the terminal today, and decent analytics doesn't really work well in terminal-based products.
[25:26] And so that's a problem for me because the people I ultimately want to serve are not in this room.
[25:30] They're not technical.
[25:31] I want DEX for non-technical people.
[25:33] I use my husband as a litmus.
[25:35] If he can work it, then we're solved.
[25:37] And that's why desktop mobile apps are coming out hopefully next week to about 500 people.
[25:42] But right now I'm flying blind.
[25:44] And up until now, I've had an awful lot of lovely qualitative feedback around DEX from people I've looked up to for years, product high priests like the lovely Martin we saw on stage earlier.
[25:53] Unbelievable, right?
[25:54] Hair on the arms.
[25:55] Humbling doesn't cover it.
[25:56] But love is not data.
[25:58] People coming back for seconds?
[26:00] I have no idea.
[26:00] Are they just being nice to me?
[26:02] Maybe.
[26:03] Which really, you know, bits actually matter.
[26:05] I just have no idea.
[26:07] But if I can't prove all these evenings and weekends are worth it, I am definitely divorce material.
[26:12] So let me show you what excites me the most.
[26:14] I want you to bring it back to something I showed you earlier, the truths.
[26:18] It ties into Martin's presentation.
[26:19] One of the truths that we're increasingly betting on a pendo is that software teams are going to operate at a radically faster pace.
[26:27] Code will change constantly.
[26:29] Humans and AI agents will increasingly build alongside each other.
[26:32] And if instrumation, if instrumentation still has to be manually maintained every time the product changes, it will drift almost immediately, right?
[26:40] So the traditional analyzer and feedback loop simply won't keep up.
[26:45] So we had to ask ourselves the same question I'm asking you all.
[26:48] Is the product that we have today built for that future?
[26:51] Or do we need to place a harder bet on a different kind of product?
[26:55] What's our FIN?
[26:56] Novus is one of those bets.
[26:58] The code base becomes a source of truth, you connect it to Novus, and then that's your product context.
[27:04] That's what exists, it's how it fits together, that's what's changed.
[27:07] The analytics and the customer behavior provide the user context, what people do, where they struggle, what creates value.
[27:13] And so you bring those together, and humans and AI agents have a much richer understanding of the product that they're changing.
[27:19] And that's the strategic bet that if software is going to be built differently for us, the products that help companies understand and improve software probably have to be built differently too.
[27:27] And because I'm incapable of keeping my day job and my side project apart, I installed Novis onto Dex, and this is after five minutes of connecting the GitHub repository.
[27:36] And honest to God, folks, five minutes, and not an engineer killed or bribed.
[27:40] No one involved apart from me and just connecting it.
[27:42] And it already worked out the product was for.
[27:44] It worked out the ICP, just looking at the code.
[27:47] It mapped the structure, automatically tagged all the pages and events, and identified important journeys you see behind me, such as the start my day ritual within DEX.
[27:54] And I hadn't even told it that that journey existed.
[27:56] And so remember two slides ago, I admitted I was flying blind.
[28:00] This starts to change out for me.
[28:01] The code base tells NOAA's what the product is, and the behavior tells how people are experiencing it.
[28:06] You put those two together and you have a living context that can help me, and increasingly these crazy agents that are building alongside me to make better decisions about what to do next.
[28:15] So the building loop got fast.
[28:16] Novus is our bet, is our fin on the learning loop just being as fast.
[28:22] But here's the thing about loop speed.
[28:25] Every time you remove one constraint, another one pops up like lemmings.
[28:29] Mine moved again last week into continuous integration.
[28:34] As I race to get to the desktop and mobile beta out, I'm doing something like 200 pull requests a day.
[28:40] I'm not doing anything actually, not me at all.
[28:42] But everyone has to be tested, type checked, linted, verified.
[28:45] So suddenly my problem isn't the cost of producing the code, it's the cost of making sure that I can sleep at night, that actually the code is safe to ship.
[28:52] And I moved from GitHub Actions to Warp Bill thinking greater save some money.
[28:55] And then this week I moved on to use jog.
[28:58] And you can see the money behind me, the difference behind me from a run this morning.
[29:01] The same gate, roughly the same job, and it's 50 cents per PR down to 17.
[29:07] So this isn't an advert, by the way, for jog, it's a bottleneck advert.
[29:12] Because this is the game now.
[29:14] You remove one constraint, throughput jumps, and then the bottleneck moves somewhere else.
[29:18] So, as I mentioned earlier to you, you need to find the new constraint, you need to redesign around it and go again and go again and go again.
[29:25] You bend the system, it's not a transformation project, it's how you keep increasing that loop speed.
[29:31] And that brings me back to the three moves.
[29:35] AI is changing the economics of building software much faster than most organizations are changing around it.
[29:42] Tiny teams now do extraordinary amounts of work on radically lower cost bases, but the companies in this room already have what these new entrants desperately want.
[29:51] You have customers, you have distribution, you have trust data brands in a place in the market.
[29:57] So the opportunity is to remove the organizational drag around those advantages.
[30:02] As many here today on stage have been saying.
[30:04] So for me, it's out hunt.
[30:06] Find that signal sooner.
[30:07] Bend the system of work.
[30:09] When the bottleneck moves, redesign around it and go around again.
[30:12] Prove impact, make the learning loop as fast as the building loop, because when everyone can build the advantages and who produces the most software, it's who learns fastest about what's worth building and who can keep bending the system around that learning.
[30:26] And I genuinely think there has never been a better time to be doing this.
[30:29] Dex, by the way, as you can see behind me, it is free if you want to play with these ideas, scan the QR code, make it your own, bastardize it.
[30:36] Most people have done that.
[30:37] I don't know if you saw earlier on Dex is on screen in the top left-hand corner.
[30:40] And then Thomas was on stage, which was deeply humbling.
[30:43] So it's 10 minutes to set up a terminal, connect your tools, and desktop and mobile app for those less technical who don't want to be in terminal coming out soon.
[30:49] So heydex.ai4 slash beta.
[30:51] And finally, if any of this has got you thinking about how your own organization needs to change, come and find me afterwards.
[30:57] My job is to basically hug the market in any way I can.
[30:59] You can be a pendo customer, not a pendul customer.
[31:02] Scan that QR code for me on LinkedIn, find me there, follow me for more rants, and DM me.
[31:07] I'm Irish, I love to talk.
[31:10] And more importantly, go hunt your signal, go bend the system of your work, go prove what works.
[31:16] And I wish everyone an utterly fabulous afternoon.
[31:18] Thank you, thank you, thank you.
[31:25] Dave, thank you so much.
[31:28] That was fantastic.
[31:29] We sadly don't have time for questions as expected.
[31:33] As expected, but we have uh definitely some time to give in the bar.
[31:39] He's in the bar.
[31:40] Dave is in the bar.
[31:42] That's such a stereo time.
[31:43] Find the Irish buyer in the car.
[31:44] I'll be at the bar, be holding up the bar.
[31:46] All right.
[31:46] But in any case find Dave and please give him again a roaring can you give me a word that is better like a different like an like synonym I always an almighty roar or an almighty roar and a stomping and like send him off with a great applause.
[32:01] Thank you thank you so much so about the next one so I think I have probably known Felix for for eight nine years so back in the days I was working um in interim position at sixth as a product manager and he was believe it or not my agile coach and I've worked with my fair share of agile coaches but nobody was like Felix because Felix also a developer like me by trade he was always a product person and he was always thinking about not really or how could can I do a retro with the team but rather like what do I need to do to change the organization and transform it and honestly we we talk in unregular intervals and I think every time I talk to Felix like we we I jump off the call I'm thinking oh yeah that's actually good I need to give that a try so he always inspires me and he will definitely do the same for you today so please welcome on stage Felix Gelha so hey everybody.