The Signal Layer: What to Build When Anything Can Be Built — Lena Hall, Akamai

AI Engineer · 19 min · 210 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:01[music]
00:12How is the conference for all of you so far?
00:16Great.
00:16Awesome.
00:18Um well, I think this was the best most productive year for so many of us.
00:25I'm Lena.
00:26A few days ago, I solved a production incident on a trail near a waterfall.
00:32My friend ran 18 agents while riding his bike.
00:37We're literally drowning in abundance.
00:40We have more output, more speed, more leverage than any of us have ever had.
00:47So, why do we have this feeling like the ground underneath is moving too fast?
00:54One of the engineers that I met at this conference said yesterday that it feels like the opportunity cost for not working 9:00 a.m. to 9:00 p.m. 6 days a week is too high right now.
01:07So, we're all token maxing.
01:09We're all working all the time.
01:12But the same abundance that made you fast, it also made everyone else fast.
01:17So, now everyone can build everything.
01:21Your competitor can build your feature this afternoon, too.
01:25So, the cost of the average just went to zero and so did its value.
01:31A year ago, the superpower, as we were told, was to be good at using AI.
01:37But 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.
01:49Cuz AI gives everyone the same answer because everybody is asking the same question.
01:55It It on data and data is a record of what has already happened.
02:00So, when you point AI at tasks like, "Tell me what users want.
02:04Make more money.
02:05What should we build?
02:07Make this viral."
02:08It answers from the common knowledge.
02:11Very competent, very confident, but also very identical to what it tells your competitor.
02:17To 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.
02:28AI is a really smart convergence machine.
02:32So, if you leave it alone, it makes everything the same.
02:37There is one decision, though, that AI can't and shouldn't make for you.
02:41It is to decide what to point at.
02:44So, the job, the new job for everyone of us is deciding what it makes, being the reason the right people choose your version over the identical-looking rest.
02:55But also, I'm sure many of you uh walked around the Expo Hall at this conference, and there are so many amazing products, so many tools and vendors.
03:04They're all solving important problems.
03:07But what Why do they all sound the same?
03:11So, when anyone can build anything, what makes me different?
03:15What makes you different?
03:16Why should anyone pick your version, your product?
03:20Um I call this work uh signal layer.
03:24And there are two There's two halves to solving it, to getting this right.
03:28So, that's how we will walk through it.
03:31The first half is knowing your signal, being able to define it very clearly.
03:37What you're building and why it's yours and not the average.
03:40So, that's the build side, the code, the product, the road map.
03:45And the second half is emitting that signal without distortion.
03:50So, making sure that your customers um making sure what your customers come to believe about you actually matches what you believe and what you've built.
04:00That's the ship side the content and go to market engineering.
04:05And I've had an unusual vantage point in this.
04:08I've built products as an engineer.
04:10I created my own as a founder.
04:12I brought other people's products to market.
04:15So three very different jobs with one identical challenge.
04:19The signal doesn't always survive.
04:22So let's start with the build side.
04:25So what do we even work on?
04:27Everything is implementable.
04:30Two years ago the best autonomous coding agents you know solved on the fraction of the tasks on the standard software benchmark and now the best agents are in the high eighties.
04:42So we nearly tripled the amount of writing and shipping barely moved a third.
04:49The benchmark was measuring the part of software engineering that has a greater and shipping is where all the ungraded parts come back in.
04:59So here is the rule underneath it.
05:03Anything that you can measure you can train against as Sarah Guo puts it.
05:09A compiler is a free grader.
05:11A test suite is a free grader.
05:14And the instant a task can grade itself you can grind a model against you know that grade until it wins.
05:22Automation of code was first because it's the most checkable thing that we have.
05:28So 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.
05:40So the model will build whatever you point it at but it will tell you nothing about where to point.
05:46Anything visible is replicatable.
05:50So now, some people when they hear everything is implementable, they panic.
05:54Um but we can flip the question.
05:56The pointing is actually the job.
05:59It has always been the job.
06:01We just had so much implementation work in the way um that we never had to get good at it.
06:08So, how do you decide where to point that?
06:11Paul Graham shared some wisdom on this.
06:14Where you find something that people genuinely want is by feeling the need yourself.
06:20Build something you and your friends need because the market hasn't formed yet, surveys can't see it, and your own need is the only signal that isn't a crap signal.
06:31And the best ideas may sound genuinely lame at first, like a guy uh strapped with a with a camera on his head live streaming his his life.
06:41That sounds really ridiculous, but it became Twitch.
06:44Um and the convergence machine doesn't really, you know, propose proactively these uh weird, specific, genuinely embarrassing ideas.
06:54But even with the Twitch example, it worked, but there were a thousand other similar startups I start startup ideas that didn't.
07:03So, the weird specific signal is necessary, but it is not sufficient.
07:10It's also really tempting to say that we just need to have good judgment and good taste and call it safe.
07:17But taste is really just preference under feedback, and preference under feedback is exactly what these systems can learn.
07:26Anything you can demonstrate enough times uh with a better or worse signal attached, the machine can eventually imitate.
07:34So, broad good taste is not really a differentiator.
07:39What actually resists training is more narrow and more durable.
07:44So, two things.
07:46Taste and judgement about what hasn't happened yet.
07:50Because there's no data for an event that hasn't occurred.
07:53And then taste and judgement embedded in a relationship that the model can't observe.
07:58What this customer in this situation with this history that you share actually needs.
08:05The model has read everything ever written about your customer, but it has never actually met them.
08:12So, if broad judgement is not safe, and the AI just handed everyone the ability to build anything, what's left to be good at?
08:21Richard Hamming uh spent his career studying why some scientists did great work and others who were just as smart didn't.
08:32He found that the great ones worked on important problems.
08:37And the problem isn't important because it just sounds impressive.
08:42It's important when you have a reasonable attack on it.
08:45For example, time travel is consequential, he would say, but it's not important because nobody has an attack on it.
08:52So, Hamming would tell you to keep 10 or 20 ideas um on important problems in the back of your mind so that when you finally have an attack, a new tool, a new angle
09:04I think that only you noticed, then you go for it.
09:08But in Hamming's world, the rarest thing was having an attack.
09:12And AI just handed everyone an attack on everything.
09:15So, the rarest thing is knowing which problem is actually worth attacking.
09:19And that judgement 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.
09:32And you don't actually need to be first.
09:34You just need to be genuinely close to a problem you actually understand where your insight is in the delta between what AI has been trained on and what should exist.
09:45So, let's say you did it.
09:46You found that sweet spot problem that the one that you had an honest attack on, that you built this thing.
09:53It's genuinely good.
09:54It's genuinely yours, not the average.
09:58You can still lose uh because knowing your signal is only half the job.
10:03The other half is getting it from your head into the head of a person that it was meant for.
10:09And it's about reaching the right people.
10:12And what do most of us do for that?
10:15We make content.
10:16So, let's talk about what uh AI convergence machine does to that.
10:23What happened to the internet in the last 2 years?
10:26Open any feed, everything has started to sound the same.
10:30The same LinkedIn posts, the same, you know, three bullet points and a bold takeaway, and the same blog post that uh says nothing but actually looks very polished.
10:42Um your readers can now pattern match AI in just half a second.
10:46So, if a model could have written your post from a one-line prompt, your reader brain just skips it for the same reason.
10:56So, AI has really learned the algorithm.
10:58It has learned the format that performs.
11:00It has learned what gets clicks, and everyone wants to hand the machine a paragraph and say, you know, "Make it viral.
11:07Make me rich."
11:09It fills every gap that you leave with sameness.
11:13So, what do you put in and what do you let it fill in?
11:17Cuz these there there are two different ways to use this thing, and they look very identical from the outside.
11:24One is you give it an average prompt, and gives you the average output.
11:28And you ship one more indistinguishable drop into an ocean of indistinguishable drops.
11:34So, you've automated your own irrelevance very efficiently.
11:38And two, you can bring in 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,
11:49the formatting, the drafting, the algorithm optimization, the cleanup around the core that it could have never generated.
11:58The signal distorts on the way out.
12:00So, you can have the signal perfectly clear for you and still watch it fall apart between your brain and your users' understanding of it.
12:10And in my experience, it breaks in three places.
12:14And there are fixes for each, but they're different depending on product, the type, and the size of the company.
12:22One of them is source distortion, which is very common in startups.
12:26Founders, actually, they usually know the signal so well that they always have this accidental gift of compressing it past legibility.
12:37They often assume the context that the audience doesn't have, and the room hears something technically very cool, but they doesn't they don't really understand why it matters.
12:48I helped this one YC company with uh this exact thing recently.
12:53Absolutely brilliant founders, genuinely new product, but every pitch that they started um was you know starting with architecture, with the clever parts, with things that they were very proud of.
13:05But it really landed as noise because the customer pain has been deleted from the whole story.
13:12So, we rewrote the opening to include the thing that the user hated, and this product actually killed.
13:18So, the same product, the same week, the next conversations converted into pilots, and then we turned that into repeatable GTM system.
13:28Organization distortion is another type of distortion that almost every big company has.
13:35As signal travels through layers of management, through legal, through sales, through every department, at every hand handoff, it gets rewound towards the average.
13:46And this really doesn't come from incompetence, it comes from investment.
13:50So, hand a founder and the person three layers down the same task and the same AI, and you get two different things.
13:59Um the founder really sweats the unaverageable details because the outcome is really theirs and they're personally invested and affected by it.
14:09And others just ship it to spec, they close Jira tickets, they were asked for, you know, something like compliance, not as much conviction.
14:18So, a long delegation chain plus convergence machine is really a factory for automating the signal out of your own company.
14:27So, the first instinct usually is to add more process, which adds layers, bureaucracy, and slows everything down.
14:35And we don't want that.
14:37Um 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,
14:48where its only job is to validate and carry the original intent across the handoffs intact.
14:55Machine distortion is another way you can lose signal.
14:59You write one careful launch, your claim, your evidence, and your scope is very clear, but then of course AI remixes it um into a tweet, into a sales deck, into a partner one-pager.
15:11For 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.
15:23So, we see the same through line.
15:25Your signal has to survive the trip undistorted.
15:30And this is something you can engineer.
15:33So, 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.
15:43Say you're building a monitoring tool.
15:45There are 12 other tools in this category, but yours does something different.
15:50It tells you what not to wake up for, for example.
15:54It stays quiet on the noise, so when you get paged at night, you believe it.
16:00So, that quiet, that trust earned by silence is your signal.
16:05So, first, say it in one sentence with the limit built in.
16:09Definitely don't say intelligent AI-native observability platform.
16:14Say something like uh stays quiet on anything it can't tie to a real user impact and shows you everything it silenced so you can overrule it.
16:24The promise and the scope are welded together here.
16:28Then make sure that the limit can't be edited out.
16:30So, in the product, every suppressed alert is visible.
16:34In the launch, statements like 90% fewer pages uh live next to statements like every silence is visible and reversible.
16:43So, when AI chops your launch into a tweet, it can keep the impressive number, but also remove the part that keeps the part that uh keeps the product honest.
16:55And before you scale it, check what people actually heard.
16:58So, give a readme to an SRE who has never seen your project and ask a person to describe the product back to you.
17:07The gap between what they say and what you meant is the distortion that you were about to broadcast.
17:14And 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.
17:23So, step back and ask what all of this, the building, the shipping, the undistorted signal, is actually for.
17:31It'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.
17:42So, that's trust.
17:44Trust is the one thing that's left with no greater.
17:47There's no benchmark for it, no reward signal.
17:50It can't be entirely automated because it's granted slowly through relationship with consent.
17:57For example, doctors who open one particular tool every morning, they didn't have that habit trained into them.
18:05And what happens if we get this wrong?
18:08Getting your signal wrong is actually not neutral.
18:11It's negative.
18:12Producing averageness is not free.
18:15You actually pay for it in tokens, in infra, in the salaried hours of good people, you know, with with customers that take a look at your product once, decide once, and never come back.
18:27So, every generic post teaches them that your name isn't worth the click.
18:33So, you spend real money to make yourself harder to choose.
18:37So, back to the main question.
18:39We got faster, but the speed is not where the value went.
18:44Uh the value moved up to deciding what is worth building, what is worth saying, what deserves trust.
18:50And where does the thing that you actually um that that you meant survives the trip to the people that it was for.
18:58So, you don't need to be first.
18:59You need a real problem and enough conviction to carry the signal clearly through to, you know, right people to find it.
19:08So, when you can build anything, you should build trust.
19:12Have the strongest conviction, define the signal yourself, protect it from distortion, and use AI aggressively for everything else.
19:21Thank you.
19:21Let's connect and happy to chat with you afterwards.
19:24Thank you.
19:26[applause]
19:40[music]