Everybody Can Cook: Guillermo Rauch on v0, Vercel and the Collapse of Specialization

Rauch introduces Vercel as the platform behind fast, innovative websites and the home of Next.js, the open-source React framework that he says powers products like Claude, Grok and Midjourney. Lenny notes that in his reader survey v0 ranked just below Cursor and GitHub among engineering tools. Rauch calls Vercel's work with v0 a way to help more people participate in building software and to make the outputs as refined as the best sites on the internet.

He frames it as the continuation of a career spent making hard things easy for developers, from Socket.IO's real-time engine to Next.js. Counting roughly five million React developers and 20 million JavaScript developers, he guesses there are at least 100 million people with aspirations to build products, a number he borrows from Slack's user base. For him the opportunity was letting those people "yap into the computer" and see a prototype or a full product appear.

The genesis came when ChatGPT launched and turned out to be good at JavaScript, Tailwind and Next.js. Vercel's tools were already in the internet's training data, so its long-term open-source bet paid off. Rauch says v0 started on OpenAI models, perhaps even before ChatGPT's release, and today uses a pipeline of specialist models that hand off work to one another.

Rauch says over 1.3 million users have interacted with v0 and that it had its largest day ever just before the interview. Vercel is now among the largest customers of most cloud providers and is hitting the limits of GPU and LLM infrastructure. A month-old feature called v0 Community already has 20,000 submissions, some with thousands of forks, which helps users who face a blank prompt box.

He sees this as the next evolution of GitHub's "social coding" idea: GitHub democratized software but you still needed to code, whereas he is after "social product building." He also describes v0 as inverting the Git commit. Engineers normally work first and summarize their intent afterward, while in v0 the intent comes first and the code is the output. A feature that produces a Git commit as a side effect was due within days.

Asked how product teams will change, Rauch's biggest prediction is that conversations between builders and customers will be mediated by live artifacts instead of abstractions. He wants people to live inside the product and tune it, rather than feeling powerless to change something when departments hand work to one another. Vercel's experience with design engineers, who have both skills, helped inspire v0, and he imagines designers shipping fully baked products and PMs prototyping into production.

On whether to learn to code, he says knowing how things work is the most important skill. He once hired specialists to translate Figma designs into CSS layouts, and he sees such tasks as translation, the kind of problem that LLM techniques lineage from Google Translate solved. His own test is to rebuild his website in v0 with each new model: it took 10 to 15 prompts last time and two on the newest, and the model even produced more accessible code than he had.

Still, he stresses that naming things precisely, such as CSS, layout and styles, lets you steer the model. Engineers who memorize every CSS property won't need that depth, which he thinks frees minds for more ambitious work. Lenny pushes on the point and Rauch agrees that conceptual understanding of systems and databases will stay valuable.

Rauch has five kids and says he pushes them on math as the ability to understand the logic behind things, and on eloquence. He references the "shape rotators versus wordcels" meme, saying Silicon Valley has historically rewarded the math side but respects both. His example is a Vercel designer who used the word "turbulence" to prompt an animated triangle in v0, which shows how vocabulary translates to style. Prompt enhancement, he says, cannot replace your own thinking or creativity.

He also takes his kids to hackathons so they can watch people present ideas. When producing software gets cheaper, he argues, you need to build an audience, learn to talk to people and develop your own signature style. If they want to code, he suggests starting with JavaScript, which he calls the English of programming languages, and notes Ramp's Super Bowl ad drove a 43X traffic increase handled by an engineer who only needed to learn Next.js.

He believes foundational infrastructure will remain a durable engineering skill, because LLMs orchestrate compilers, clouds and frameworks rather than writing them from scratch, given context and compute limits. At Vercel, with 600 people and 150 engineers, some of the best v0 creations come from marketing, sales and product. He describes a PM speccing a feature as a live PRD with error, success and slow-stream states, and says the team will not be told to simply "just ship it."

Lenny asks whether v0 can move beyond prototypes. Rauch says yes: an enterprise customer works only in v0, someone sold a first website to a client built entirely with it, and an investor told him two friends got engaged using a v0 site. Because v0 is integrated with Vercel's serverless infrastructure, the path from prompt, screenshot or pasted PRD to something served to a huge audience is end to end.

His first-use tips are to be as ambitious as you want, steer toward inspiration, fork from the community if you lack ideas, and suspend disbelief about what the tool knows. He suggests focusing on what the end user should experience rather than prescribing implementation. He also advises treating it like an agency or an engineer you coach: when stuck, he has often unblocked it by simply saying "try something else."

Rauch keeps a personal test of finding something v0 cannot build. On a flight from Tokyo to San Francisco with bad internet, he built his own flight tracker with a flight data API after finding existing apps bloated with ads. He wasn't prescriptive: v0 chose Mapbox and Leaflet, and he began with design, then made it real, working experience first as Vercel itself was built.

When too many planes slowed the app, he told v0 to improve performance and it produced a canvas-based overlay that renders tens of thousands of flights. The whole thing took under two hours on his $20 subscription, versus weeks of engineering, or a couple of days from Vercel's most cracked engineer. Rauch says he was the bottleneck, not the engineer, and that v0 even accounted for the curvature of the earth when asked to draw a dashed line between airports.

He relays a user's description of v0 as a "super genius five-year-old PhD with ADHD," and adds he does not want to oversell it. He thinks the term AGI undersells what is being built, since the tools already show sparks of superhuman ability.

Lenny suggests three core skills for building with AI: deciding what to build, making it look good and getting it unstuck. Rauch adds coaching and connects it to React's design principle of the escape hatch, a way out when the component model doesn't fit, with deliberately scary names like `dangerouslySetInnerHTML`. He says React's component model also helps LLMs, since components are reusable building blocks.

In v0 the escape hatch is the generated Next.js code, which can be edited or reviewed by experts. He tells how a Luma Labs design engineer got stuck on a v0 chat of about 120 iterations, copied the code into ChatGPT o1 and got a solution. Rauch connects this to a Facebook principle of getting unblocked by seeking help, and says AIs can now be that help, even in a hybrid project with Git.

In a live demo, Rauch prompts a contact sales form styled after a clothing brand, using the enhance-prompt feature to fight writer's block and showing the model's visible reasoning, a feature inspired by DeepSeek. The output uses shadcn/ui, Tailwind and React, the same component system he says Grok uses. He then asks for a more serious Charles Schwab look and notes the model now acts as an editor, tweaking what exists, which scales to large projects.

He forks a Luma Labs recruiting site with about 2,000 forks, then pastes a screenshot of fortune.com to generate a news site and uses another site's sepia style as a second screenshot. A precise inline prompt centers an element, and he describes v0 as "code last rather than code first." It applied a CSS filter he did not know existed, which is why he says you can't be too opinionated with the tool.

Rauch cautions that screenshots are not for cloning other people's sites but for using your own art boards, designs and Slack posts. Lenny says such work would take an engineer days or weeks, to which Rauch adds months or maybe never.

Rauch calls the DeepSeek thinking-tokens moment a big one, since users love seeing how the AI thinks and send bug reports pointing out where it went wrong. His top advice for startup founders is to create many chances for in-product feedback, as Vercel did early on with a Stripe-inspired emoji form that streamed into Slack. AI products receive a constant stream of feedback, and he says it will be hard to compete with a product that has such a tight loop.

He says Vercel open-sourced much of what makes v0 work, including templates and the AI SDK, which connects any model and can generate UI rather than text. In a demo he asks an open-source chat app for San Francisco's weather and gets a component back, which he calls generative UI. Lenny raises vertical AI tools, and Rauch says he is convinced expert tools are the future, citing chatprd.com, OpenEvidence and Vercel's legal team's tool, built by a lawyer.

Asked whether AI could one day build a custom Harvey for a lawyer in real time, Rauch says eventually yes, but humans will want guardrails. Companies that build such tools also put their opinion and credibility behind them, and he sees AI as a copilot and partner rather than something that runs our lives.

Asked how to build taste, Rauch says it is a skill, not an inborn trait. Vercel has an internal operating principle of increasing exposure hours: quantify how much time you spend watching people use your products and other products. He advises trying lots of products, shipping, not being shy of self-promotion and seeing how people react.

He gives the example of Grok's chat interface, which makes Enter create a new line on mobile and submit on desktop, because that matches how people expect to behave. You have to pay attention to such details and decide what you want to see in the world. He warns that people overrate how well their products work and recommends handing your product to someone and watching them use it.

Lenny asks for a number, and Rauch says more is always better because the inertia is to stay inside your own head. He colors his calendar for team one-on-ones and customer meetings, tries customers' products, and aims for a third of his meetings to be customer sessions. Vercel also invites customers to demo live to the executive team or company, which always reveals pain points.

On limitations, Rauch says v0 can import code via zip files, with Git coming soon, and can do full stack work and connect to APIs. Marketplace integrations for databases and AI models were about to land. Errors are still possible, but v0 is agentic and often fixes its own. Very large code bases remain hard for AI, so he suggests scoping tasks to a component or file, which suits React and Next.js file-based routing.

For design tips, he tells how a bank CIO was unhappy with a music-festival brochure design until Rauch suggested simply giving v0 feedback such as "make it more jazzy, make it pop." He recommends using style tokens like neobrutalist, minimalist, newspaper-like or vintage. On product quality, he says great products are a thousand little details, and compares each feature to adopting a puppy, so the best builders say nine nos for every yes.

Inside Vercel, designers use video models for conference animations and marketing builds interactive infrastructure demos, and a ritual called Demo Fridays gives people permission to step out of their lane. Rauch ends by saying AI is becoming synonymous with software, and that his vision, captured by a San Francisco billboard reading "Everybody Can Cook," is a future where, if you can dream it, you can ship it.