---
title: "Intelligence on tap: build today what everyone gets later"
description: An AI app that costs $1,000 a month is enterprise software today, and the same app becomes a consumer product once AI gets a thousand times cheaper. What I took from Garry Tan and Mark Pincus, the founder of Zynga, on building it now.
date: 2026-11-04
tags: AI, consumer products, AI costs, startups, Y Combinator
---

Garry Tan, who runs Y Combinator (YC), says we're moving to intelligence "on tap", like water from the tap. Mark Pincus, the founder of Zynga, takes up the image and adds that we're not there yet. So today, an AI app that costs $1,000 a month is enterprise software, even when it's made for people at home. Their advice is to build it now and be ready when the price falls.

I got to this part of the interview through one of my own agents. I sent it the YouTube link on Telegram, set at minute 37, and asked what the tap water and the consumer and enterprise talk meant. A minute later, it sent back a clear summary of the passage. It got two details wrong: the interviewer's name, and who first said "on tap".

The video is [Zynga Founder: Consumer Is Not Investible Right Now](https://www.youtube.com/watch?v=oHwUD9b9_pg), an episode of YC's Main Function series from June. YC funds and coaches young startups. Mark's new book is *Life at the Speed of Play*. This note covers the last ten minutes, where they talk about price, timing and free.

## A $1,000 app is enterprise software

The same app can be enterprise or consumer. What decides it is the price, not who the app is for.

Mark's example is an AI that listens. Earlier in the episode, he says he walks around talking to his AI. He wants it to listen in on a conversation, like "a smart other person at the table". Garry calls that a consumer company worth 10 or 100 billion dollars that "no one's built yet".

Then Garry gives the price. "You can have that app now. It'll just cost $1,000 a month." He, Mark and a bunch of their friends could pay that. "That's not consumer. That's basically enterprise."

He also measures the gap. Since what he calls the "Opus 4.5 moment" in December, real work with these models costs "tens to hundreds of thousands of dollars". So the right moment for consumer is "three orders of magnitude away": a thousand times cheaper. His guess for the consumer revolution is 2029. Mark's answer: "That sounds right."

A thousand times cheaper turns $1,000 a month into $1. That's my arithmetic, not theirs, but it's the step from a few friends to everyone.

The asking half of that app is already here: it's what I did with this video. The listening half, all day long, is the part with the $1,000 price.

<figure class="fd-fig tw-price" aria-labelledby="tw1-cap" data-desc="The same always-listening AI app at two prices. Today it costs about 1,000 dollars a month: Garry, Mark and a few of their friends can pay, drawn as four yellow squares, so it is enterprise software. A thousand times cheaper, it costs about 1 dollar a month: anyone can pay, drawn as forty squares with the same four yellow ones among them, so it is a consumer product. Underneath: Garry's guess for the consumer revolution is 2029.">
<div class="mp-two-row">
<div class="mp-two-card">
<p class="mp-two-who">Today</p>
<p class="mp-two-name">$1,000 a month</p>
<p class="tw-ppl" aria-hidden="true"><i class="tw-me"></i><i class="tw-me"></i><i class="tw-me"></i><i class="tw-me"></i></p>
<p class="mp-two-line">Garry, Mark and a few friends can pay</p>
<p class="mp-two-line tw-kind">Enterprise</p>
</div>
<p class="mp-two-vs">then</p>
<div class="mp-two-card mp-two-peter">
<p class="mp-two-who">1,000 times cheaper</p>
<p class="mp-two-name">$1 a month</p>
<p class="tw-ppl" aria-hidden="true"><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i class="tw-me"></i><i></i><i></i><i></i><i></i><i></i><i></i><i class="tw-me"></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i class="tw-me"></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i class="tw-me"></i><i></i><i></i><i></i></p>
<p class="mp-two-line">Anyone can pay</p>
<p class="mp-two-line tw-kind">Consumer</p>
</div>
</div>
<p class="mp-strip"><b>Garry's guess</b><span>The consumer revolution comes in 2029.</span></p>
<figcaption id="tw1-cap">Their $1,000 a month and "three orders of magnitude". The $1 is my arithmetic from the two.</figcaption>
</figure>

## We are the time travellers

If you can pay for tomorrow's AI today, you live a few years ahead of everyone else. Garry and Mark see that as a head start.

Earlier, Garry had joked that neither of them was a time traveller. Then he took it back: "Actually, we are time travellers right now." Most people can't spend that much, so they can't have it yet. Garry says that pains him. Mark offers the other way to see it: everyone will have this soon.

So Mark says to "figure out the primitives right now", the basic building blocks. Garry adds that you can start building now and work backward from the cheap future. Mark's test question is simple. How will all of these services work "when that compute is basically free"?

The hard part is the wait, and Mark has lived it once. In 1999, he built an enterprise software company, support.com, while investors only wanted consumer websites. Then the consumer internet crashed. eBay took off early, but Amazon took longer. In the fall of 2002, Amazon announced a big quarter, at a time when the internet "was supposed to be over". That's when he thought: "Oh, this is finally happening." It was about six years in.

Garry's version is harsher: you have to be wrong "for like almost a decade". Mark adds that then everyone feels late, "and they are". Today, investors push the other way. Garry tells of a startup in his office hours with some of the best consumer numbers he'd seen in years. Its investors told it to pivot to enterprise.

<figure class="fd-fig tw-time" aria-labelledby="tw2-cap" data-desc="Two timelines side by side. Then, the internet: a telecom boom that laid mostly dark fiber, consumer hype and a crash, eBay taking off early, and in the fall of 2002 Amazon's big quarter, about six years in. Now, AI: the Opus 4.5 moment in December, real work costing tens to hundreds of thousands of dollars, investors telling a consumer startup to pivot to enterprise, and consumer three orders of magnitude away, with Garry's guess of 2029. Underneath: Garry says you have to be wrong for almost a decade.">
<div class="mp-two-row">
<div class="mp-two-card">
<p class="mp-two-who">Then: the internet</p>
<p class="mp-two-name">About 6 years</p>
<p class="mp-two-line">A telecom boom, mostly dark fiber</p>
<p class="mp-two-line">Consumer hype, then the crash</p>
<p class="mp-two-line">eBay takes off early</p>
<p class="mp-two-line tw-kind">Fall 2002: Amazon's big quarter</p>
</div>
<p class="mp-two-vs">vs</p>
<div class="mp-two-card mp-two-peter">
<p class="mp-two-who">Now: AI</p>
<p class="mp-two-name">Guess: 2029</p>
<p class="mp-two-line">December: the Opus 4.5 moment</p>
<p class="mp-two-line">Real work costs tens to hundreds of thousands of dollars</p>
<p class="mp-two-line">Investors say: pivot to enterprise</p>
<p class="mp-two-line tw-kind">Consumer: 1,000 times cheaper</p>
</div>
</div>
<p class="mp-strip"><b>Garry</b><span>"You have to be wrong for like almost a decade."</span></p>
<figcaption id="tw2-cap">Mark's memory of the internet and Garry's reading of AI today, as they tell them. The 2029 is a guess.</figcaption>
</figure>

## The iPhone was a cost curve first

You could have predicted the iPhone from a few falling prices. Garry says you can do the same with AI today.

He calls it one of the classic ways to find a startup idea. Take a log graph of a cost curve, or of what a technology can do. "You could have guessed the moment that the iPhone could have been possible." Memory got cheaper, displays got cheaper and computing got cheaper. At some point, a person could afford the whole phone. "And so you can do this right now with the cost of intelligence."

The episode doesn't draw the AI curve, but it gives two hints of how steep it is. Garry quotes his friend Suhail Doshi: one thing we can count on is "a 10,000x increase in inference", the work a model does each time it answers. Garry ties the figure to Jeff Dean, Google's chief scientist. And Elon Musk says games will one day be drawn pixel by pixel by AI, while you look around. Mark finds that "very expensive" for now. But "you can't bet against Elon": he's probably right, at a later point in time.

<figure class="fd-fig tw-curve" aria-labelledby="tw3-cap" data-desc="A schematic cost curve on a log scale. The cost of one AI product falls in a straight line over time. A dashed line marks what a person will pay. A yellow square moves down the curve under a big label in three steps: too expensive for anyone, then enterprise while only a few can pay, then consumer once the cost drops under the dashed line. Underneath: the iPhone came when memory, displays and computing fell under that line, and Garry says to read the cost of intelligence the same way.">
<p class="ea-now ea-say3" aria-hidden="true"><span>Too expensive</span><span>Enterprise: a few can pay</span><span>Consumer: anyone can pay</span></p>
<div class="tw-plot">
<svg class="tw-line" viewBox="0 0 100 100" preserveAspectRatio="none" aria-hidden="true"><path d="M0 8 L100 92"/></svg>
<p class="tw-pay" aria-hidden="true"></p>
<p class="tw-zone tw-zone-hi">Enterprise</p>
<p class="tw-zone tw-zone-lo">Consumer</p>
<i class="tw-dot" aria-hidden="true"></i>
</div>
<p class="tw-axis">Cost, log scale, over time</p>
<p class="tw-key"><span class="tw-k-cost">The cost of one product</span><span class="tw-k-pay">What a person will pay</span></p>
<p class="mp-strip"><b>The iPhone</b><span>Memory, displays and computing fell under the line.</span></p>
<p class="mp-strip"><b>AI now</b><span>The same graph, with the cost of intelligence.</span></p>
<figcaption id="tw3-cap">Schematic. Nobody in the episode gives the slope or a date for the crossing.</figcaption>
</figure>

## Free has won twice already

Mark's plan for consumer is what he calls "the business plan of free". He says it's one of the rules of the internet: "anything that can be free will be free."

His first company, Freeloader, gave away an interactive screensaver. It competed with the Flying Toasters from Berkeley Systems, which sold for $35. "It's not even new. It's just better."

Then came Zynga. The games industry was worth $23 billion, with one model: you went to a store and bought a box for $60. Zynga asked if decent games could be free, and free worked again.

So his question for AI is the same. Take any consumer service and imagine it with free, unlimited AI inside. Garry finishes the thought: "That's when we'll see the new Meta. That's when we'll see the new Snap."

Mark also sees room. Half of his phone's home screen is empty, he says, and the other half holds generic apps: clock, notes, photos, camera, weather. Apart from ChatGPT and Claude, no consumer app grabs his attention, and certainly no game. Few services are ones we can't imagine life without, like Google. That's why he thinks the chance to build a new one "has never been greater".

<figure class="fd-fig tw-free" aria-labelledby="tw4-cap" data-desc="Three markets, paid against free. Screensavers: the Flying Toasters sold for 35 dollars, and Freeloader gave an interactive one away. Games: a 60-dollar box at the store, in a 23-billion-dollar industry, against Zynga's free social games. AI services: about 1,000 dollars a month today, against free and unlimited AI, which doesn't exist yet. Underneath, Mark's rule: anything that can be free will be free.">
<div class="tm-table">
<p class="tm-row tm-head"><span>Market</span><span>Paid</span><span>Free</span></p>
<p class="tm-row"><b>Screensavers</b><span>Flying Toasters, $35</span><span>Freeloader, interactive</span></p>
<p class="tm-row"><b>Games</b><span>A $60 box, in a $23B industry</span><span>Zynga's social games</span></p>
<p class="tm-row"><b>AI services</b><span>About $1,000 a month today</span><span class="tw-next">Free and unlimited: not yet</span></p>
</div>
<p class="mp-strip"><b>Mark's rule</b><span>"Anything that can be free will be free."</span></p>
<figcaption id="tw4-cap">His two companies and the paid products they beat, as he tells it. The last row is the bet.</figcaption>
</figure>

## Try it

The whole stretch runs about ten minutes, from minute 32. If you only have three, start at "Intelligence on tap".

- [Zynga Founder: Consumer Is Not Investible Right Now](https://www.youtube.com/watch?v=oHwUD9b9_pg), Y Combinator, June 2026. [The AI at the table at 7:00](https://www.youtube.com/watch?v=oHwUD9b9_pg&t=420s), [intelligence on tap at 37:04](https://www.youtube.com/watch?v=oHwUD9b9_pg&t=2224s), [the business plan of free at 38:41](https://www.youtube.com/watch?v=oHwUD9b9_pg&t=2321s).
- Mark's book, [Life at the Speed of Play](https://www.lifeatthespeedofplay.com/).
- Two of my notes on what AI costs: [spending more tokens](/notes/tokenmaxxing-fat-skills-real-tests.html), from an earlier episode with Garry, and [counting what tokens buy](/notes/tokenomics-dont-count-tokens.html).

Pick one service you use every week. Ask Mark's question: how would it work with free, unlimited AI inside? Then build the smallest piece of it for yourself, at today's price. If you try it, tell me which service you picked.