How is the conference for all of you so far? Great, awesome. Well, I think this was the best, most productive year for so many of us. I'm Lina. A few days ago, I solved a production incident on a trail near a waterfall. My friend ran 18 agents while riding his bike. We're literally drowning in abundance. We have more output, more speed, more leverage than any of us have ever had. So why do we have this feeling that the ground underneath is moving too fast? One of the engineers that I met at this conference said yesterday that it feels like the opportunity cost for not working 9 a.m. to 9 p.m., six days a week, is too high right now. So we're all token maxing. We're all working all the time. But the same abundance that made you fast also made everyone else fast. So now everyone can build everything. Your competitor can build your feature this afternoon too. So the cost of the average just went to zero, and so did its value.
A year ago, the superpower, as we were told, was to be good at using AI. But models got so good, and they got so easy, and everybody now is a lot more skilled at using AI. And everybody's pointing AI at the same goals. AI gives everyone the same answer because everybody is asking the same question. It runs on data, and data is a record of what has already happened. So when you point AI at tasks like, tell me what users want, make more money, what should we build, make this viral, it answers from the common knowledge. Very competent, very confident, but also very identical to what it tells your competitor. To see something that data doesn't show yet, we need to have a vision, a point of view, a read on where it's going, and then use all that automation to execute it. AI is a really smart convergence machine. So if you leave it alone, it makes everything the same.
There is one decision, though, that AI can't and shouldn't make for you. It is to decide what to point at. So the new job for every one of us is deciding what it makes, being the reason the right people choose your version over the identical-looking vest. But also, I'm sure many of you walked around the expo hall at this conference, and there are so many amazing products, so many tools and vendors. They're all solving important problems. But why do they all sound the same?
So when anyone can build anything, what makes me different? What makes you different? Why should anyone pick your version, your product? I call this work a signal layer. And there are two halves to solving it and to getting this right. So that's how we will walk through it. The first half is knowing your signal, being able to define it very clearly, what you're building and why it's yours and not the average. So that's the build side, the code, the product, the roadmap. And the second half is emitting that signal without distortion, so making sure what your customers come to believe about you actually matches what you believe and what you've built. That's the ship side, the content and go-to-market engineering.
And I've had an unusual vantage point in this. I've built products as an engineer, I created my own as a founder, I brought other people's products to market. So, three very different jobs with one identical challenge. The signal doesn't always survive. So let's start with the build side. What do we even work on? Everything is implementable. Two years ago, the best autonomous coding agents solved only a fraction of the tasks on the standard software benchmark. And now the best agents are in the high 80s. So we nearly tripled the amount of writing, and shipping barely moved a third. The benchmark was measuring the part of software engineering that has a grader. And shipping is where all the ungraded parts come back in. So here is the rule underneath it. Anything that you can measure, you can train against, as Sarah Gould puts it. A compiler is a free grader. A test suite is a free grader. And the instant a task can grade itself, you can grind the model against that grade until it wins. Automation of code was first because it's the most checkable thing that we have. So implementation is converging for free for everyone at the same time. And the most buildable thing and the most valuable thing are almost never the same thing. So the model will build whatever you pointed at, but it will tell you nothing about where to point. Anything visible is replicatable.
Now, some people, when they hear everything is implementable, they panic. But we can flip the question. The pointing is actually the job. It has always been the job. We just had so much implementation work in the way that we never had to get good at it. So how do you decide where to point? Paul Graham shared some wisdom on this. The way you find something that people genuinely want is by feeling the need yourself. Build something you and your friends need because the market hasn't formed yet, service can't see it, and your own need is the only signal that isn't a crap signal. And the best ideas may sound genuinely lame at first, like a guy strapped with a camera on his head, live-streaming his life. It sounds really ridiculous, but it became Twitch. And the convergence machine doesn't really propose proactively these weird, specific, genuinely embarrassing ideas. But even with the Twitch example, it worked, but there were a thousand other similar startup ideas that didn't. So the weird, specific signal is necessary, but it is not sufficient.
It's also really tempting to say that we just need to have good judgment and good taste and call it safe. But taste is really just preference under feedback. And preference under feedback is exactly what these systems can learn. Anything you can demonstrate enough times with a better or worse signal attached, the machine can eventually imitate. So broad, good taste is not really a differentiator. What actually resists training is more narrow and more durable. So, two things: taste and judgment about what hasn't happened yet, because there's no data for an event that hasn't occurred. And then taste and judgment embedded in a relationship that the model can't observe. What this customer in this situation, with this history that you share, actually needs. The model has read everything ever written about your customer, but it has never actually met them. So broad judgment is not safe.
And if AI just handed everyone the ability to build anything, what's left to be good at? Richard Hamming spent his career studying why some scientists did great work and others who were just as smart didn't. He found that the great ones worked on important problems. And the problem isn't important because it just sounds impressive. It's important when you have a reasonable attack on it. For example, time travel is consequential, he would say, but it's not important because nobody has an attack on it. So Hamming would tell you to keep 10 or 20 ideas on important problems in the back of your mind so that when you finally have an attack, a new tool, a new angle, a thing that only you noticed, then you go for it. But in Hamming's world, the rare thing was having an attack. And AI just handed everyone an attack on everything. So the rare thing is knowing which problem is actually worth attacking. And that judgment comes from being a real person, close to a real domain, with your own battle scars, your weirdly specific experience, the thing that you care about more than is reasonable. And you don't actually need to be first. You just need to be genuinely close to a problem. Your insight is in the delta between what AI has been trained on and what should exist.
So let's say you did it. You found that sweet-spot problem, the one that you had an honest attack on, that you built the thing. It's genuinely good, it's genuinely yours, not the average. You can still lose. Because knowing your signal is only half the job. The other half is getting it from your head into the head of a person that it was meant for. And it's about reaching the right people. And what do most of us do for that? We make content. So let's talk about what AI, the convergence machine, does to that.
What happened to the internet in the last two years? Open any feed, everything has started to sound the same. The same LinkedIn posts, the same three bullet points and a bold takeaway at the end. The same blog post that says nothing but actually looks very polished. Your readers can now pattern-match AI in just half a second. So if a model could have written your post from a one-line prompt, your reader's brain just skips it for the same reason. So AI has really learned the algorithm. It has learned the format that performs. It has learned what gets clicks. And everyone wants to hand the machine a paragraph and say, make it viral, make me rich. It fills every gap that you leave with sameness.
So what do you put in, and what do you let it fill in? Because there are two different ways to use this thing. And they look very identical from the outside. One is you give it an average prompt, and it gives you the average output. And you ship one more indistinguishable drop into an ocean of indistinguishable drops. So you've automated your own irrelevance very efficiently. And two, you can bring it the part that it can't have: your specific point of view, the real story that you were actually in the room for, and then let the machine do the converging work, the formatting, the drafting, the algorithm optimization, the cleanup around the core that it could have never generated.
The signal distorts on the way out. So you can have the signal perfectly clear for you and still watch it fall apart between your brain and your users' understanding of it. And in my experience, it breaks in three places. And there are fixes for each, but they're different depending on the product, the type, and the size of the company.
One of them is source distortion, which is very common in startups. Founders, actually, usually know the signal so well that they always have this accidental gift of compressing it past legibility. They often assume the context that the audience doesn't have, and the room hears something technically very cool, but they don't really understand why it matters. I helped this one YC company with this exact thing recently. Absolutely brilliant founders, genuinely new product, but every pitch that they started was starting with architecture, with the clever parts, with things that they were very proud of. But it really landed as noise because the customer pain had been deleted from the whole story. So we rewrote the opening to include the thing that the user hated and this product actually killed. So the same product, the same week, the next conversations converted into pilots, and then we turned that into a repeatable GTM system.
Organization distortion is another type of distortion that almost every big company has. As signal travels through layers of management, through legal, through sales, through every department, at every handoff, it gets re-rounded toward the average. And this really doesn't come from incompetence. It comes from investment. So hand a founder and the person three layers down the same task and the same AI, and you get two different things. The founder really sweats the un-averageable details because the outcome is really theirs. And they're personally invested and affected by it. And others just ship it to spec. They close Jira tickets. They were asked for something like compliance, not as much conviction. So a long delegation chain plus convergence machine is really a factory for automating the signal out of your own company. So the first instinct usually is to add more process, which adds layers, bureaucracy, and slows everything down. And we don't want that. To fix this, we have to help take the signal back and reattach it to the outcome like a founder, and add a very thin signal layer to your go-to-market engineering, where its only job is to validate and carry the original intent across the handoffs intact.
Machine distortion is another way you can lose signal. You write one careful launch. Your claim, your evidence, and your scope are very clear. But then, of course, AI remixes it into a tweet, into a sales deck, into a partner one-pager. For example, you might have had this one very narrow eval that scored 94%. But it was repeated enough times that your customers actually heard it as a promise. So we see the same through line. Your signal has to survive the trip undistorted. And this is something you can engineer.
So we need a thin signal layer, a small deliberate function whose job is to make sure that what your users take away is still the specific thing you meant. Say you're building a monitoring tool. There are 12 other tools in this category. But yours does something different. It tells you what not to wake up for, for example. It stays quiet on the noise so when you get paged at night, you believe it. So that quiet, that trust earned by silence, is your signal. So first, say it in one sentence with the limit built in. Definitely don't say intelligent AI-native observability platform. Say something like, stays quiet on anything it can't tie to a real user impact and shows you everything it silenced so you can overrule it. The promise and the scope are welded together here.
Then make sure that the limit can't be edited out. So in the product, every suppressed alert is visible. In the launch, statements like 90% fewer pages live next to statements like every silence is visible and reversible. So when AI chops your launch into a tweet, it can keep the impressive number but also remove the part that keeps the product honest. And before you scale it, check what people actually heard. So give a readme to an SRE who has never seen your project and ask that person to describe the product back to you. The gap between what they say and what you meant is the distortion that you were about to broadcast. And it's a very lightweight signal layer. And a lot of it is buildable. So you can automate more of the checking and the catching and the surveying than most people realize.
So step back and ask what all of this, the building, the shipping, the undistorted signal, is actually for. It's for one thing: getting a human, or increasingly an agent, to choose you and rely on you when they have infinite identical-looking alternatives. So that's trust. Trust is the one thing that's left with no grader. There is no benchmark for it, no reward signal. It can't be entirely automated because it's granted slowly through a relationship with consent. For example, doctors who open one particular tool every morning, they didn't have that habit trained into them.
And what happens if we get this wrong? Getting your signal wrong is actually not neutral. It's negative. Producing averageness is not free. You actually pay for it in tokens, in infra, in the salaried hours of good people, with customers that take a look at your product once, decide once, and never come back. So every generic post teaches them that your name isn't worth the click. So you spend real money to make yourself harder to choose.
So back to the main question. We got faster. But the speed is not where the value went. The value moved up to deciding what is worth building, what is worth saying, what deserves trust, and where the thing that you meant survives the trip to the people that it was for. So you don't need to be first. You need a real problem and enough conviction to carry the signal clearly through to the right people to find it. So when you can build anything, you should build trust, have the strongest conviction, define the signal yourself, protect it from distortion, and use AI aggressively for everything else. Thank you. Let's connect, and happy to chat with you afterwards. Thank you.
And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it.
And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it.
And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. ... ... ... you find something that people genuinely want is by feeling the need yourself. Build something you and your friends need because the market hasn't formed yet, service can't see it, and your own need is the only signal that isn't a crap signal. And the best ideas may sound genuinely lame at first, like a guy,
strapped with a camera on his head, live streaming his life. It sounds really ridiculous, but it became Twitch. And the convergence machine doesn't really, you know, propose proactively these weird, specific, genuinely embarrassing ideas. But even with the Twitch example, it worked, but there were a thousand other similar startup startup ideas that didn't. So the weird, specific signal is necessary, but it is not sufficient. It's also really tempting to say that we just need to have good judgment and good taste and call it safe. But taste is really just preference under feedback. And preference under feedback is exactly
what these systems can learn. Anything you can demonstrate enough times with a better or worse signal attached, the machine can eventually imitate. So broad, good taste is not really a differentiator. What actually resists training is more narrow and more durable. So two things, taste and judgment about what hasn't happened yet because there's no data for an event that hasn't occurred. And then taste and judgment embedded in a relationship that the model can't observe. What this customer in this situation, with this history that you share actually needs. The model has read everything ever written about your
customer, but it has never actually met them. So broad judgment is not safe. And the AI just handed everyone the ability to build anything what's left to be good at. Richard Hamming spent his career studying why some scientists did great work and others who were just as smart didn't. He found that the great ones worked on important problems. And the problem isn't important because it just sounds impressive. It's important when you have a reasonable attack on it. For example, time travel is consequential. He would say, but it's not important because nobody has an attack on it. So Hemming would tell you to keep 10 or 20 ideas on important problems
in the back of your mind so that when you finally have an attack, a new tool, a new angle, a thing that only you noticed, then you go for it. But in Hemming's world, the rare thing was having an attack. And AI just handed everyone an attack on everything. So the rare thing is knowing which problem is actually worth attacking. And that judgment comes from being a real person, close to a real domain, with your own battle scars, your weirdly specific experience, the thing that you care about more than is reasonable. And you don't actually need to be
first. You just need to be genuinely close to a problem. You actually understand where your insight is in the delta between what AI has to do. What you have been trained on and what should exist. So let's say you did it. You found that sweet spot problem, the one that you had an honest attack on, that you built the thing. It's genuinely good, it's genuinely yours, not the average. You can still lose. Because knowing your signal is only half the job. The other half is getting it from your head into the head of a person that it was meant for.
And it's about reaching the right people. And what do most of us do for that? We make content. So let's talk about what AI convergence machine does to that. What happened to the internet in the last two years? Open any feed, everything has started to sound the same. The same LinkedIn posts, the same, you know, three bullet points and a bold takeaway at the end. The same blog post that says nothing but actually looks very polished. Your readers can now pattern match AI in just half a second. So if a model could have written your post from a one line prompt, your reader brain just skips it for the same reason.
So AI has really learned the algorithm. It has learned the format that performs. It has learned what gets clicks. And everyone wants to hand the machine a paragraph and say, you know, make it viral, make me rich. It fills every gap that you leave with sameness. So what do you put in and what do you let it fill in? Because there are two different ways to use this thing. And they look very identical from the outside. One is you give it an average prompt and gives you the average output. And you ship one more indistinguishable drop into an ocean of indistinguishable drops. So you've automated your own irrelevance very efficiently.
And two, you can bring it the part that it can't have. Your specific point of view, the real story that you were actually in the room for, and then let the machine do the converging work. The formatting, the drafting, the algorithm optimization, the cleanup around the core that it could have never generated. The signal distorts on the way out. So you can have the signal perfectly clear for you and still watch it fall apart between your brain and your users' understanding of it. And in my experience, it breaks in three places. And there are fixes for each, but they're different depending on the product, the type, and the size of the company.
One of them is source distortion, which is very common in startups. Founders, actually, they usually know the signal so well that they always have this accidental gift of compressing it past legibility. They often assume the context that the audience doesn't have, and the room hears something technically very cool, but they don't really understand why it matters. I helped this one YC company with this exact thing recently. Absolutely brilliant founders, genuinely new product, but every pitch that they started was, you know, starting with architecture, with the clever parts, with things that they were very proud of.
But it really landed as noise because the customer pain has been deleted from the whole story. So we rewrote the opening to include the thing that the user hated and this product actually killed. So the same product, the same week, the next conversations converted into pilots, and then we turned that into repeatable GTM system. Organization distortion is another type of distortion that almost every big company has. As signal travels through layers of management, through legal, through sales, through every department, at every handoff, it gets re-rounded towards the average. And this really doesn't come from incompetence. It comes from investment.
So hand a founder and the person three layers down the same task and the same AI, and you get two different things. The founder really sweats the un-averageable details because the outcome is really theirs. And they're personally invested and affected by it. And others just ship it to spec. They close Jira tickets. They were asked for something like compliance, not as much conviction. So a long delegation chain plus convergence machine is really a factory for automating the signal out of your own company. So the first instinct usually is to add more process, which adds layers, bureaucracy, and slows everything down.
And we don't want that. To fix this, we have to help take the signal back and reattach it to the outcome like a founder, and add the very thin signal layer to your go-to-market engineering, where its only job is to validate and carry the original intent across the handoffs intact. Machine distortion is another way you can lose signal. You write one careful launch, your claim, your evidence, and your scope is very clear. But then of course, AI remixes it into a tweet, into a sales deck, into a partner one-pager. For example, you might have had this one very narrow eval that scored 94%. But it was repeated enough times that your customers actually heard it as a promise.
So we see the same through line. Your signal has to survive the trip undistorted. And this is something you can engineer. So we need a thin signal layer, a small deliberate function, whose job is to make sure that what your users take away is still the specific thing you meant. Say you're building a monitoring tool. There are 12 other tools in this category. But yours does something different. It tells you what not to wake up for, for example. It stays quiet on the noise so when you get paged at night, you believe it. So that quiet, that trust earned by silence is your signal. So first, say it in one sentence with the limit built in.
Definitely don't say intelligent AI-native observability platform. Say something like, stays quiet on anything it can't tie to a real user impact and shows you everything it silenced so you can overrule it. The promise and the scope are welded together here. Then make sure that the limit can't be edited out. So in the product, every suppressed alert is visible. In the launch, statements like 90% fewer pages live next to statements like every silence is visible and reversible. So when AI chops your launch into a tweet, it can keep the impressive number but also remove the part that keeps the product honest. And before you scale it, check what people actually heard.
So give a read me to an SRE who has never seen your project and ask a person to describe the product back to you. The gap between what they say and what you meant is the distortion that you were about to broadcast. And it's a very lightweight signal layer. And a lot of it is buildable. So you can automate more of the checking and the catching and the surveying than most people realize. So step back and ask what all of this, the building, the shipping, the undistorted signal is actually for. It's for one thing of getting a human or increasingly an agent to choose you and rely on you when they have an infinite identical looking alternatives. So that's trust.
Trust is the one thing that's left with no greater. There is no benchmark for it, no reward signal. It can't be entirely automated because it's granted slowly through a relationship with consent. For example, doctors who open one particular tool every morning, they didn't have that habit trained into them. And what happens if we get this wrong? Getting your signal wrong is actually not neutral. It's negative. Producing averageness is not free. You actually pay for it in tokens, in infra, in the salaried hours of good people, with customers that take a look at your product once, decide once and never come back.
So every generic post teaches them that your name isn't worth the click. So you spend real money to make yourself harder to choose. So back to the main question. We got faster. But the speed is not where the value went. The value moved up to deciding what is worth building, what is worth saying, what deserves trust. And where the thing that you actually, that you meant, survives the trip to the people that it was for. So you don't need to be first. You need a real problem and enough conviction to carry the signal clearly through to, you know, right people to find it.
So when you can build anything, you should build trust, have the strongest conviction, define the signal yourself, protect it from distortion, and use AI aggressively for everything else. Thank you. Let's connect and happy to chat with you afterwards. Thank you. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it.
And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. And I look forward to it. ... ... ...