Guillermo Rauch, founder of Vercel: Developer Experience, AI and v0
Guillermo Rauch says developer experience on its own is a dangerous fitness function: a DevTool has two customers, and the experience of the person using what your developer builds matters even more.
Guillermo Rauch is the founder of Vercel, creator of Next.js and Socket.IO, a prolific DevTools investor, and one of the people most responsible for making modern web software faster to build and deploy.
We talk about staying close to the product while building a 500-person company, hiring for growth and full-stack range, putting design before code in v0, the compounding value of integrations, how to balance ambition with readiness, and why every software company is becoming an AI company.
The Q&A below is experimental — extracted from the episode transcript with AI. If you spot a mistake, let us know at jack@scalingdevtools.com.
What does your role look like now that Vercel is a large company?
I still use and talk to the products, often on weekends or holidays, because I need to know whether they motivate me to create and whether I would recommend them to people I know. At the same time, Vercel the company is itself a product: I think about employee onboarding, the path from an idea to something shipped, and how to make the organization an invention machine that can keep creating products such as Next.js.
What do you look for when hiring?
I look at the slope of a person's growth: are they learning quickly, hungry, visible and shipping? Communication matters more as AI makes the physical crafting of things easier, because people must communicate both with models and with customers and colleagues. I value people who can span customer discovery, engineering, design, presentation and business judgment even when their formal role will not include every one of those jobs.
What changes when software becomes more fluid through AI?
Software used to feel fixed and expensive to translate from intent into data structures, libraries and code. Vercel first made deployment fluid; AI now delegates much of that translation too. It feels strange to give up handcrafting every line, much as automatic formatting initially did, but Prettier freed teams from pointless formatting debates. AI can similarly free people to think about more important parts of the creation while a human still directs the orchestra.
Who is v0 for, and how is it different from a code-first AI editor?
An editor such as Cursor begins with a repository and assumes comfort with code. v0 is code-last: it still produces visible, auditable code, but starts with the design, the idea and the desired experience. That extends the design-engineer model — somebody with aesthetic judgment who can also ship — to many more people. Users understand speed, intuition and quality even when they do not understand the implementation, so design-led development starts from the end customer's reality.
Why is developer experience alone the wrong thing to optimize?
A DevTool has two customers: the developer using it and the person using what that developer creates. A responsive, pleasurable development system matters, but if it is optimized only for its direct user it may win early affection without becoming economically sustainable. The end user's outcome is ultimately more important, so work backward from that customer as well.
Why have integrations mattered so much to Vercel?
The more we polished our integrations, the more the business grew. AI-generated software will often be a thin layer of interface over a large sea of existing capabilities, because neither people nor models should recreate every library and system. An additive product with a precise point of view is easier to adopt than a demand to rip out something a company has optimized for decades. If you are not replacing everything, good integration is how your value enters the existing system.
How do you decide whether an idea should become a Vercel feature, product or platform?
I use the hierarchy of feature, product and platform. A new company first has to prove it can do one thing excellently; a platform must also expose primitives and extension points that let others build. We did choose product-building mode for the AI SDK and v0 because AI changed the foundation itself. That work also dogfoods Vercel and strengthens the larger platform, but beginning any new product should feel consequential — endless expansion is how companies ship slop.
How should DevTools founders approach large companies experimenting with AI?
New AI-native startups and experimental teams inside enterprises can be excellent early customers, but a pilot at a giant company does not automatically become the whole enterprise. As the product matures, larger and more critical workloads become possible. The trap is building hopes around an experimental mandate before the technology and customer are ready. Partner with those teams, learn, and prepare for the later moment when an enterprise wants to leap forward safely.
What does it look like to say no to a huge customer until you are ready?
A founder once replied to Auth0's CTO, who wanted to run everything on the product, by saying: we are not ready; message me in a few months. That was advanced founder judgment. Vercel has often moved large customers incrementally, sometimes one page at a time, so both sides can prove ROI and reliability. Founders need ambition and realism at once: dream of the large outcome while honestly delivering exceptional value at today's level of maturity.
What is your broadest advice to a DevTools founder?
Keep the proposition simple and put the product directly inside its presentation. Browserbase lets a visitor immediately use a virtual browser; LiveKit lets them speak with a real-time agent. Marketing, product education and documentation become one experience, with no delay between arriving and understanding the value. A page can load instantly while still having a two-month communication latency if nobody can work out what the product does.
Links & references
This episode is brought to you by WorkOS.