LLenny's Podcast
← All episodes
Guillermo Rauch (founder and CEO of Vercel, creators of v0 a13 April 2025

Everyone’s an engineer now: Inside v0’s mission to create a hundred million builders

6Frameworks
15Insights

Frameworks in this episode

Peak Performance3 steps

Exposure Hours

Build taste as a trainable skill by quantifying time spent watching real people use products

Read →
Self-Mastery3 steps

Learn the Tokens, Not the Depth

In the AI era, master the symbolic vocabulary of a domain rather than its exhaustive depth

Read →
Strategy3 steps

Nine Nos For Every Yes

Guard product quality with creative restraint — every feature you say yes to is a puppy you must care for forever

Read →
Productivity4 steps

Product-First AI Prompting

Steer AI builders by describing the end-user experience ambitiously, then iterate like you're coaching a collaborator

Read →
Innovation3 steps

The Escape Hatch Principle

Abstractions should let power users drop to raw control when the model doesn't fit their problem

Read →
Entrepreneurship3 steps

The In-Product Feedback Loop

Stream user reactions straight into your team's consciousness by building feedback into the product itself

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 4

Hot Take20:30

Many Programming Specializations Are Just Translation Tasks

Rauch argues that many programming specializations, like converting a Figma design into styled CSS/Tailwind code, are fundamentally translation tasks. He points out the transformer architecture behind LLMs traces back to machine translation systems like Google Translate, which is why models cross the chasm so well. When he last regenerated his own website in v0 it took two prompts, versus 10-15 with a prior model.

  • Styling a design into layout/CSS code used to be a hired specialist role
  • The transformer architecture descends from generative translation systems like Google Translate
  • Regenerating his own site went from 10-15 prompts to 2 prompts across model generations
  • v0's output followed web accessibility standards better than Rauch's own hand-written code

“a lot of the programming jobs to be done that used to be specializations that I think are going away in a way or the…”

Guillermo Rauch · 21:00
#ai-coding#careers#translation#css
Hot Take32:00

The Engineers AI Won't Replace: Foundational Infrastructure Builders

Rauch frames LLMs as oracles that write software but are bounded by context windows and compute, so an agent can't write a cloud, framework, and compiler from scratch. The model orchestrates existing tools and infrastructure rather than rebuilding the universe. Engineers who build foundational infrastructure will therefore stay extremely empowered for years or decades.

  • LLMs are bounded by context windows and computational constraints
  • Models orchestrate existing infrastructure rather than writing compilers or clouds from scratch
  • Foundational infrastructure engineering is a durable skill for decades to come
  • The idea of ChatGPT building the next ChatGPT is, in Rauch's view, a long way off if ever

“the engineers that learn foundational infrastructure are probably going to be extremely empowered still for years to come.”

Guillermo Rauch · 33:00

“In order to create an apple, you have to create the entire underlying universe.”

Guillermo Rauch · 33:00
#ai#infrastructure#careers#engineering
Hot Take1:04:30

Why Rauch Believes Expert Vertical AI Tools Will Win

Rauch is convinced that specialized, expert AI tools are the future, giving examples like ChatPRD for PRDs, OpenEvidence for doctors, and getGC.ai for legal (whose CEO is herself a lawyer). Even though people could ask ChatGPT the same questions, a founder who obsesses over one vertical, keeps it up to date, and puts their reputation behind it will be hard to compete with. He argues humans will always want guardrails and a point of view, not an AI left entirely to its own devices.

  • Expert AI tools like ChatPRD, OpenEvidence, and getGC.ai target specific verticals
  • A founder obsessed with one problem beats a general model even when the model 'could' answer
  • Domain-expert founders lend direction, opinion, and reputation to the tool
  • Humans will still want guardrails; AI remains a copilot, not something running our lives

“I'm absolutely convinced that expert AI tools are the future.”

Guillermo Rauch · 1:05:00
#ai#startups#vertical-saas#product
Hot Take1:25:00

AI Will Become Synonymous With Software

Rauch predicts a future where AI simply becomes synonymous with software, so much so that we should stop talking about 'AI' as separate. At Vercel, everyone can ship: designers generate animations with video models, marketing builds interactive infrastructure demos in v0, and one designer's interactive explainer of Vercel's compute went viral. He frames Vercel as a software company that builds software and uses software to build software.

  • AI is becoming embedded everywhere and will be synonymous with software
  • Designers, marketing, and sales engineers at Vercel all ship using v0 and AI models
  • A designer's interactive infrastructure explainer outperformed any static diagram and went viral
  • Vercel's 'demo Fridays' ritual gives people permission to step outside their job and build

“I just see a future where AI becomes synonymous with software.”

Guillermo Rauch · 1:25:00

“AI equals software now and we're a software company.”

Guillermo Rauch · 1:25:30
#ai#future#culture#vercel

Explainer· 3

Explainer11:30

v0's Scale and Why Rauch Calls It the Next GitHub

Rauch shares that over 1.3 million users have interacted with v0, that it is now one of the largest customers of most cloud and GPU providers, and that the month-old v0 community feature already has 20,000 submissions. Users fork each other's apps (some with thousands of forks) and remix them, which Rauch frames as 'social product building' and the next evolution of GitHub.

  • Over 1.3 million users have interacted with v0
  • v0 is hitting the limits of GPU and LLM infrastructure worldwide
  • The v0 community had 20,000 submissions in under a month, with apps forked 1,200 to 6,000 times
  • Rauch sees it as 'social product building' where GitHub was 'social coding'

“over 1.3 million users have interacted with Vzero so far.”

Guillermo Rauch · 11:30

“In many ways, I see this as the next evolution of GitHub.”

Guillermo Rauch · 13:30
#v0#community#github#open-source
Explainer14:30

How v0 Inverts the Git Commit

Rauch describes a fundamental workflow inversion. An engineer normally does the work first, then writes a git commit summarizing their intent afterward. With v0 you start with the intent, in plain English, and the code is the output. The git commit becomes a side effect the tool can generate for you.

  • Traditionally engineers write the commit message after finishing the work to summarize intent
  • v0 flips this: you state intent first and code is the output
  • The git commit can be auto-generated as a side effect

“D0 inverts that. The git commit is you go into the chat and say, "Please change the color of this button and when I click…”

Guillermo Rauch · 15:00
#v0#workflow#git#ai-coding
Explainer1:08:00

Taste Is a Skill You Build Through Exposure Hours

Rauch rejects the idea that taste is innate. He treats it as a developable skill built through 'exposure hours,' one of Vercel's internal operating principles: quantify how much time you spend watching people use your product and others'. He stresses giving your product to real users and watching them struggle, because you tend to overrate how well your own product works.

  • Taste is a skill you develop, not something you're born with
  • Vercel treats 'increasing exposure hours' as an internal operating principle
  • Watch how people use your product and competitors' products to build the muscle
  • You overrate your own product; watching real users exposes you to the pain of reality

“At Versel, we have one of our sort of internal operating principles as increasing exposure hours.”

Guillermo Rauch · 1:08:30

“I see it as a skill that you can develop”

Guillermo Rauch · 1:09:00
#taste#design#product#vercel

Story· 2

Story08:30

Where the '100 Million Builders' Number Came From

Rauch explains how v0's target market was sized on the back of a napkin. There are maybe 5 million React developers and 20 million JavaScript developers, but he pegs the addressable pool of would-be product builders at 100 million by borrowing Slack's monthly active user count. The logic: those 100 million people already go into Slack every day to talk about the products they wish existed.

  • Roughly 5 million React developers and 20 million JavaScript developers exist today
  • Rauch estimates 100 million people have aspirations to build products
  • The number is anchored to Slack's ~100 million monthly active users
  • v0's mission is to expand the total addressable market of people who can ship real products

“Slack has about a 100 million monthly active users. And what you do on Slack is you go in it and you talk to people.”

Guillermo Rauch · 08:30
#v0#market-sizing#product-builders#vercel
Story39:00

Building a Better Flight Radar Mid-Flight for $20

Bored on a Tokyo to San Francisco flight with terrible internet, Rauch decided to build his own flight tracker in v0 because existing ones were bloated and full of ads. Working experience-first, he told v0 to render planes on a map, then to handle the performance problem of too many flights (it chose a canvas overlay rendering tens of thousands of planes), then to make it full-stack against a flight API. It took under two hours and cost his $20 monthly subscription.

  • Built during a flight with the worst possible internet connection
  • v0 chose Mapbox, Leaflet, and a canvas-based overlay without being told to
  • The result rendered tens of thousands of flights and handled earth-curvature math for dashed flight lines
  • Took under two hours; an engineer would have needed weeks and tens of thousands of dollars

“that one probably took less than two hours with the worst internet.”

Guillermo Rauch · 42:00

“Because I'm the bottleneck, not the engineer.”

Guillermo Rauch · 42:30
#v0#story#demo#productivity

Q&A· 1

Q&A24:30

What Rauch Is Teaching His Five Kids for an AI Future

Asked how far kids should go into software engineering, Rauch (a father of five) says he pushes math hard as the foundation of logical reasoning, and prizes eloquence and linguistic ability equally. He references the 'word cells vs shape rotators' meme, arguing both sides of the brain matter, and stresses presenting and putting yourself out there, taking his kids to hackathons to build that muscle.

  • Math is treated as a fundamental, non-negotiable skill
  • Eloquence and vocabulary let you steer models toward the right references (e.g. the word 'turbulence' for an animation)
  • He cites the 'word cells versus shape rotators' meme and values both
  • Presenting and building an audience matters as the marginal cost of making software drops

“if you don't know math really well, you're out of my house. Just kidding.”

Guillermo Rauch · 24:30

“So developing eloquence and your linguistic ability I think is going to be very important.”

Guillermo Rauch · 27:30
#education#careers#kids#skills

Takeaway· 5

Takeaway22:30

The Skill That Survives: Knowing How Things Work

Even as translation tasks are automated, Rauch says knowing how things work under the hood stays valuable because it lets you steer the model with precise vocabulary. Knowing the right tokens, like CSS, layout, box model, lets you influence the model to follow your intention. The tradeoff is you can go into each subject with less encyclopedic depth.

  • Knowing the symbolic systems (the right tokens) lets you direct the model precisely
  • You will need less deep, encyclopedic knowledge of each subject
  • Rauch's engineers who memorized every CSS property probably won't need that depth in the future
  • Freeing up that mental space lets you tackle more ambitious problems

“the TLDDR would be knowing how things work, the symbolic systems, and that will mean that you have to probably go into each subject with…”

Guillermo Rauch · 23:00
#ai-coding#skills#careers#prompting
Takeaway31:30

Why Rauch Says Learn JavaScript First

Rauch calls JavaScript the 'English of programming languages' because learning it lets you reach billions of devices. He notes it's no coincidence that ChatGPT, Claude, and Gemini all reach for JavaScript and React when building web apps. He cites Vercel customer Ramp, whose engineer only needed to learn Next.js before their Super Bowl ad drove a 43x traffic spike with 100% uptime.

  • JavaScript reaches billions of devices, making it the lingua franca of the web
  • Major LLMs default to JavaScript and React when generating web apps
  • Vercel is described as 'AWS in easy mode' for people who want scale without configuring the cloud
  • Ramp's engineer only needed Next.js; their Super Bowl ad drove a 43x traffic increase with no downtime

“JavaScript is in my mind has always been almost like the English of programming languages. It's a language that if you learn it, you reach…”

Guillermo Rauch · 31:30
#javascript#careers#vercel#nextjs
Takeaway46:30

Escape Hatches and Getting Unstuck With Another AI

Rauch praises React's 'escape hatch' design principle, where the framework hands control back to the developer when it can't model a problem (even an API literally named 'dangerously set inner HTML'). v0's escape hatch is that you can see and edit the generated code. He shares that a Luma Labs engineer, stuck after a hundred v0 iterations, copied the generated code into OpenAI's o1, which solved it, illustrating that you can now get unstuck by seeking help from other AIs.

  • React's 'escape hatch' hands control back to the developer when the framework can't model a problem
  • v0's escape hatch is full visibility into and editability of the generated code
  • A stuck Luma Labs engineer pasted v0's code into o1 to get unblocked
  • Facebook's principle 'don't get blocked, seek to get unblocked' now extends to seeking help from other AIs

“one of these escape hatches is called dangerously set inner HTML.”

Guillermo Rauch · 47:30

“you can seek help from other AIs to get unstuck”

Guillermo Rauch · 49:00
#v0#react#debugging#ai-coding
Takeaway1:01:30

Rauch's #1 Advice for Founders: Build Feedback Into the Product

Rauch's top advice to any startup founder is to create many opportunities for users to give feedback inside the product itself. Drawing inspiration from Stripe, early Vercel had a slick inline feedback button with four emoji that piped user sentiment straight into Slack. He argues AI products have such a tight feedback loop that it will be hard for non-AI products to compete.

  • Create lots of in-product opportunities for users to give feedback
  • Early Vercel copied Stripe's inline feedback form (four emoji) that fed straight into Slack
  • AI products generate a constant stream of thumbs-up/down feedback
  • That tight feedback loop improves the next model, prompt, fine-tune, and RAG iteration

“create a lot of opportunities for people to give you feedback inside the product.”

Guillermo Rauch · 1:01:30
#product#founders#feedback#startups
Takeaway1:20:30

Nine Nos for Every Yes: A Feature Is Like Adopting a Puppy

On product quality, Rauch says the best product builders say nine nos for every yes, because saying yes to a feature is like adopting a puppy: it grows into a demanding beast you have to care for. Great products are made of a thousand little details and are never truly done, so creative restraint is essential. He cites Midjourney's mobile web experience as proof that obsessive love and restraint produce great results.

  • The best product builders say nine nos for every yes
  • A new feature is like a puppy that grows into a demanding beast
  • Great products are a thousand little details and are never finished
  • Creative restraint plus dogfooding (as with Midjourney's mobile web) produces quality

“the best product builders will say nine nos for every yes because when you say yes is like adopting a puppy.”

Guillermo Rauch · 1:21:00

“A feature is like adopting a puppy.”

Guillermo Rauch · 1:21:00
#product#quality#focus#craft