Emmett Shear

The consumer internet window is open again — because of AI

Emmett Shear argues that for the first time in 5 to 7 years, building a consumer internet company is credibly a good idea again, because AI lets founders reimagine consumer experiences from the ground up.

“For the first time in maybe 5 to 7 years, it feels like credibly trying to start a consumer internet company, like the ones that, like, I was so excited to start in 2007, is, like, potentially a good idea. And that's because of AI.”

You weren't born uncreative — you installed a 'don't say dumb things' sensor

Shear claims every kid can generate ideas endlessly; adults lose it by installing an internal censor that kills bad ideas before they're voiced. The more pressure you put on yourself not to say dumb things, the more your idea generator gets disrupted.

“I think most 10-year-olds and definitely most 5-year-olds are capable of generating ideas for what to do about something or to like play pretend almost indefinitely. Like they don't run out of ideas. It's as you get older, somehow you, what you learn to do is you learn to stomp down the ideas that are like bad and to not say dumb things. But the more pressure you put on yourself not to say dumb things, the more your inner idea generator, it like gets disrupted.”

Steal thisTo brainstorm, deliberately disable the inner censor and let yourself say obviously bad ideas — most will be bad, but don't stop there.

Have you tried solving the problem by solving the problem?

Shear's life-motto heuristic: people facing an obvious, solvable problem instinctively look for hacks, workarounds, or ways to dodge it. The reminder is to just attack the root problem directly — but only when it's genuinely solvable and worth solving.

“And one of my favorite ones of those that has become almost like a life motto is like, have you tried solving the problem by solving the problem? And that sounds dumb, right? Like that sounds, it's one of those like Zen koan pieces of advice that when you first hear it is like, are you serious?”

Steal thisWhen stuck, ask whether you're dodging the real problem with a hack — if it's solvable and worth it, go straight at it instead.

We give the advice we need to hear

Shear's heuristic: the advice you find yourself dispensing is often advice you yourself need right now. Whenever you give a piece of advice, check whether it's secretly meant for you.

“My version of this is we give the advice we need to hear.”

Steal thisEvery time you give advice, pause and ask: is this advice I need to hear right now?

How Twitch was born: 40 interviews to understand why anyone streams games

Shear realized after four years at Justin.tv he had no idea why people streamed video games. He did ~40 interviews with the small universe of streamers — not asking what to build, but probing why they did it — and learned the product was fundamentally about the streamers, not viewers.

“And then I had the realization, this is one of those epiphany moments where I truly saw I have no idea why anyone would stream video games. Like, I don't really want to do it. And I have all these— I could— I saw myself building products for these people for the past 4 years at Justin.tv and not really having any idea why they did the thing they did at all.”

How Twitch was born: 40 interviews to understand why anyone streams games

Shear realized after four years at Justin.tv he had no idea why people streamed video games. He did ~40 interviews with the small universe of streamers — not asking what to build, but probing why they did it — and learned the product was fundamentally about the streamers, not viewers.

“And then I had the realization, this is one of those epiphany moments where I truly saw I have no idea why anyone would stream video games. Like, I don't really want to do it. And I have all these— I could— I saw myself building products for these people for the past 4 years at Justin.tv and not really having any idea why they did the thing they did at all.”

Cut any feature that doesn't move the customer's core motivations

Once Shear learned that audience, money, and 'love' (feeling wanted) explained 98% of streamer motivation, he had a filter: features like live polls were rejected because they didn't grow audience, make money, or make streamers feel more loved — chat blowing up already did that.

“And it was really the realizing, the realization that like it was just those 3 things basically explained 98% of their motivation and we could, anything that didn't move the needle on that could be ignored. So a good example of that's like polls. Everyone would ask for polls. It seems like a cool feature. Live polls, of course. Are you going to have a bigger audience with the live polls? Not particularly. Are you going to make more money? No.”

Steal thisDistill your customer's 2-3 core motivations, then kill any requested feature that doesn't move one of them — even if it sounds cool.

Cut any feature that doesn't move the customer's core motivations

Once Shear learned that audience, money, and 'love' (feeling wanted) explained 98% of streamer motivation, he had a filter: features like live polls were rejected because they didn't grow audience, make money, or make streamers feel more loved — chat blowing up already did that.

“And it was really the realizing, the realization that like it was just those 3 things basically explained 98% of their motivation and we could, anything that didn't move the needle on that could be ignored. So a good example of that's like polls. Everyone would ask for polls. It seems like a cool feature. Live polls, of course. Are you going to have a bigger audience with the live polls? Not particularly. Are you going to make more money? No.”

Steal thisDistill your customer's 2-3 core motivations, then kill any requested feature that doesn't move one of them — even if it sounds cool.

Why consumer beats B2B for reinvention: in consumer, the experience IS the product

Shear's model: in B2B SaaS the experience isn't the product, so reimagining it rarely reopens a segment. In consumer, the experience is what you sell, so a new AI-enabled experience 100% reopens the segment — like mobile did.

“And what's, what's cool about consumer is for B2B SaaS, the experience isn't the product. And so reimagining the experience does not reopen a— it can, but it usually does not reopen a segment. In consumer, reimagining an experience 100% Reopens the segment because the thing you're selling is the experience. The thing, the reason people use your product is it's a different experience.”

Disrupt Yelp by making raw video the system of record

Shear's AI startup thesis: many consumer apps are just databases of canonical rows that users build by hand at write-time. With AI that can watch video, you could store a raw video of a meal and let the AI extract any metadata on demand — even noise level — leveling the playing field against incumbents like Yelp.

“And where I, what we think AI has opened up the possibility for is a huge inversion there. What if the thing you did, you gave us was just a video of your meal and, or, you know, photos of you, but ideally just like a video. Of your, of the meal, of you talking about the meal, of whether you had a good time or not, you and your friend shooting the shit about what did you like that one? No, I like this one. Like, and what if we just saved that video raw and then an AI watched it and extracted a cached version of that, of the metadata.”

Steal thisFind a forms-and-database consumer app, replace the structured write-time data with raw user video, and let AI extract any field on demand.

Why transformers won: they keep improving as you add data and compute

Shear's explanation of LLMs: most ML algorithms overfit fast and give diminishing returns as you add data and compute. The transformer's breakthrough is that it keeps benefiting from more compute and data, letting you train it on the entire domain of human knowledge.

“Like, most machine learning algorithms, you kind of, you overfit very fast and more processing, more data. Explain overfit.”

LLMs have high crystallized intelligence but low fluid intelligence

Shear argues today's models are overfit on all explicit human knowledge — great at recalling or interpolating between seen problems, but bad at genuinely novel reasoning (like his gears-on-a-wall flag puzzle). In psychometric terms: high crystallized intelligence, low fluid intelligence.

“if you give it a sort of the formal psychiatric psychometrics approach, it has a very high crystallized intelligence, but a pretty low fluid intelligence right now. Now that could change, but like today, that's the state of affairs.”

AI will turn everyone into a Rick Rubin

Shear predicts generative AI will de-skill the act of creating sounds but greatly up-skill curation and taste — the ability to give exact feedback and pick 'this song, not that song.' The scarce skill becomes directing infinite generated cuts, like a producer.

“AI is going to turn us all into Rick Rubins for generative AI. Like that, that skill set or the ability to have a musician come to you and help them produce their best music. That's the thing you need to be able to do because it's easy to generate 1,000 cuts, but there's infinite cuts you could generate.”

The Kasparov analogy for AI risk: you don't need to know which move beats you

Shear's mental model for AI danger: a superintelligence is like Garry Kasparov — you can be certain he'll checkmate you even though you can't name the piece he'll use. The threat isn't one specific action; it's that something is simply much smarter than you and can self-improve.

“I can tell you with confidence that Garry Kasparov is going to kick your ass at chess right now. And you ask me, well, how is he going to checkmate me? Which piece is he going to use? And I'm like, oh, I don't know. And you're like, you can't even tell me what piece he's going to use and you're saying he's going to checkmate me? You're just a pessimist. I'm like, no, no, no, you don't understand. He's better at chess than you. That means he's going to checkmate you.”

AI danger doesn't require a bad actor — just a literal-minded genie

Shear's core alignment worry: even a good person asking for a reasonable thing ('maximize this company's free cash flow') could get a world-destroying plan, because the AI solves the literal goal. The fix is building a new species that genuinely cares about the things humans care about.

“the human doesn't need a bad motivation. I think people imagine, well, humans have had powerful tools for a long time. Bad people with powerful tools have done bad things for a long time. The solution is good people with powerful tools countering them. The problem is Even if you're a good person with a powerful tool, good things to ask for, reasonable things good people would ask for”

Shear puts AI doomsday odds at 3-30%

Asked for his probability of the catastrophic AI scenario, Shear refuses a point estimate and gives a range: somewhere between 3% and 30% chance of a very bad outcome — high enough to urgently warrant action.

“So I think of it as a range of uncertainty. And I would say that the true probability, I believe, is somewhere between 3 to 30%, which—”

Why world-changers write long: 'Dunkin Theory 201, Size Does Matter'

Shear cites a blog post (he names it 'Dugin Theory 201, Size Does Matter' by Steve Yegge) arguing that people who change the world with writing all write very long posts, because you need enough time in someone's head to install your voice — paired with pithy summaries people can repeat.

“There's this great blog post, Dugin Theory 201, Size Does Matter. By Steve Yeager that's about why the people who change the world with their writing all write really long blog posts. And it's basically like you just need some amount of time in someone's head to, like, we were talking about this earlier, like to install your agent, to install the voice.”

'You know what you should do' — the Paul Graham ambition trick

Shear describes Paul Graham's signature move: he 'deludes' himself about how great your business could be and invites you into that vision, asking 'what if it goes right?' instead of staring at the hard problems — which raises the founder's ceiling and ambition.

“And from that vantage point, what you're doing is super, like, what if it does, what if it goes right is sort of what, what he invites you to ask, right? What if, stop, stop asking yourself, stop seeing all the hard problems and all the shit you're gonna have to do. Ask yourself, what if, what if what we're doing works? What if it goes right?”

Steal thisReframe a project by asking 'what if it goes right and we keep going?' to surface the biggest version of what you're building.

Bezos remembered every Twitch detail and added a new idea every meeting

Shear presented Twitch to Bezos about twice a year for his first 3-4 years at Amazon. Two things happened every time: Bezos remembered everything from the prior meeting without reviewing notes, and he'd offer at least one genuinely new idea Shear hadn't thought of — on a business Shear thought about constantly.

“First of all, he would remember everything we told him the first meeting. And I don't think he was reviewing extensive notes someone else took because I don't know when he would have the time to do that. I observed him going from meeting to meeting and he did not review notes. I think he just remembered at least the high points. And the other thing was consistently he would read our plan and he would then ask a question about why we didn't do a certain thing or he'd give us an idea for a thing we could do that I hadn't thought of before.”

Andy Jassy's criticism move: come to the founder's side of the table

Shear admires how Andy Jassy criticizes by signaling total belief in you — your talent, effort, team, and opportunity — then expressing genuine confusion about why the results aren't better, framed as 'we're in this together.' Because it's sincere, it makes you want to go fix it rather than feeling judged.

“Andy has this like ability to criticize you in a way that conveys 100%. I know that you're amazing. I know that your plan is good or, you know, like, or that you're at least capable of making a really good plan. I know that you're working really hard and I know that you are smart and you have a great team. And we have a huge opportunity, and yet somehow your results are bullshit, which must— I don't know what's wrong, but we're in this together”

Steal thisWhen giving hard feedback, lead with sincere belief in the person, then express shared confusion about the results — 'we're in this together,' not 'you failed.'

The cold-water trick that short-circuits a panic attack

Shaan shares Emmett Shear's tip: submerging your face in cold water triggers the mammalian diving reflex, which cuts the panic-attack loop. A linked demo shows a heart rate dropping from 103 to 47 bpm after ~20 seconds.

“this may sound weird, But the mammalian diving reflex is triggered by, triggered by submerging your face in cold water and cuts the panic attack loop. Can be very, very effective, but obviously hard to do in public.”

Steal thisWhen a panic attack starts, plunge your face into ice water for 10-20 seconds to trigger the diving reflex and slow your heart.

Impatience with action, patience with results; only lead bullets

Shaan shares two Twitch-era mottos he put atop every weekly memo. Emmett Shear corrected his impatience with results, so he wrote 'impatience with action, patience with results.' The second, from Ben Horowitz, is that there are no silver bullets, only lead bullets: you keep firing many things until the thing falls over.

“At the top I wrote in bold, impatience with action, patience with results. I said, that's our team motto. I'm putting it up here mostly for myself to remember impatience with action. That's when impatience is good, is when you're being impatient about taking action. But impatience is bad when you're impatient about results.”

Steal thisBe impatient about taking action but patient about results, and fire many lead bullets rather than hunting for one silver bullet.

Andrew Chen went to college at 12 and hid it until senior year

Sam recounts how a16z's Andrew Chen scored high on the SAT in 6th grade and was invited into a University of Washington program that moved gifted kids straight into college dorms. Twitch's Emmett Shear was reportedly in the same kind of program.

“when he was in about 6th grade, he took the SAT and scored really high. And when he did, I think it's the University, or University of Washington does this thing where every 5 or 10 years, uh, they take 5 or 10 students per year who are in 6th or 7th grade, sometimes younger, like 12 years old, whichever grade that is. And they asked them to come to college, to come to University of Washington. And he was one of the students.”

Emmett Shear's leadership shift: ask questions instead of saying 'this is stupid'

Shaan relays how Twitch CEO Emmett Shear changed from grilling and shooting down ideas in meetings to asking 'how did you arrive at this?' and 'what else did you consider?', assuming smart people had reasons, then giving direct feedback privately.

“now I ask, I see that you mentioned this, how did you arrive at this? Or what other solutions did you consider? And he's like, that little switch of going from like, this is dumb, why don't we do it this way? To why did you come up with this idea? And tell me, walk me through your thinking, or what other things did you consider and why did you settle on this? He's like, that changes everything.”

Steal thisWhen you think a team's plan is dumb, ask how they arrived at it and what they considered instead of overruling them, then give critical feedback one-on-one.

Kyle Vogt: built Twitch as a detour, then sold Cruise to GM for $1B

Emmett Shear's point that for technical products domain experts (not naive outsiders) disrupt best: Kyle Vogt had wanted to build self-driving cars since high school, treated Twitch as a detour, then started Cruise and sold it to GM for $1 billion roughly two years in, before self-driving cars were on the road.

“He goes, Kyle had been trying to build self-driving cars since he was in high school. Like, Twitch was like a detour for him. And he's like, you know, the timing wasn't right back then, but he had been thinking about this and working on this, fiddling with this for a long time. He was probably one of the 5 people on Earth who should have started a self-driving car company was Kyle. And so he starts Cruise, sells Cruise for $1 billion to GM”

Take technical risk, not market risk: definitely valuable, questionably possible

Shaan shares Twitch CEO Emmett Shear's framing: most founders chase low-technical-risk, questionably-valuable ideas, but the better bet is 'definitely valuable, questionably possible.' If you can build a self-driving car, demand is certain; the only risk is whether it can be done.

“He goes, there's, you know, for every one entrepreneur that's doing what Kyle did, there's probably 999, um, that are doing like the opposite risk profile. So it's like low technical risk, you could definitely, it's like definitely possible, questionably valuable. And he's like, I think more people should be doing the opposite, which is definitely valuable, questionably possible.”

Steal thisChoose problems where demand is certain and only the technical execution is in question.

Six customer calls is enough to see the pattern

Shaan relays Twitch CEO Emmett Shear's rule for customer interviews: by the sixth conversation you hear the same things over and over, so you can predict answers before they're said and don't need 50 calls. Sam adds that even at scale, calling about 10 users reveals the pattern.

“go talk to people. By the 6th conversation, you'll hear the same thing over and over again. It takes 6 phone calls basically to figure out the pattern. And by the, by the 6th one, you'll know.”

Steal thisRun customer interviews until you can predict the answers; that usually takes about six conversations, not fifty.

The three-question customer interview that built Twitch

Shaan shares Emmett Shear's entire customer-interview method while pivoting Justin.tv to Twitch: ask what they like about their current platform, what they dislike, and what it would take to switch. Then build exactly that, return in two weeks, and ask if they're ready.

“I asked 3 questions. I said, what do you like about your current platform? What do you dislike about your current platform? And what will it take to get you to switch to Twitch? And that's what he asked every single customer.”

Steal thisInterview customers with three questions: what they like, what they dislike, and what it takes to switch; then build that and circle back.