The internet, rebuilt for agents: Rauch on 60% agentic deployments at Vercel

Key takeaways from a Sourcery interview with Vercel founder and CEO Guillermo RauchDealroom has a profile for this one. Try Dealroom → (October 2026):

Agents now ship the platform. Coding agents accounted for just 3% of deployments on Vercel at the start of 2026; by the time of this interview that share had climbed to around 60%. Software is increasingly, in Rauch's words, robots telling other robots what to do.

Scale by the numbers. Vercel took roughly a decade to reach its first billion deployments, then added a further 1.5 billion in the following ten months. The interview also references the reported $600M annualised revenue, around 485,000 paying teams and the $300M Series F that valued the company at $9.3bn, with industry giants such as OpenAI and Meta among the enterprises deploying on the platform.

An index of AI. Vercel distributes AI tokens and offers every major model — OpenAI, Anthropic, open and proprietary — which Rauch casts as giving the company a unique lens on where the industry is heading: it has emerged almost as an index of the AI market.

Reject non-understanding. Rauch's viral hot take is that developers should neither blind-ship agent output nor insist on understanding every line: he still reads code, asks the AI to explain its work back to him and compares the role to an engineering manager, and has shipped an internal proof system meant to stop models shipping slop or leaking data.

Open source as moat. For more than a decade Vercel has open-sourced its tooling — Next.js is the framework nearly the whole industry uses for modern web experiences — and Rauch argues the more open information models can absorb, the better they serve the ecosystem: good karma and good business at once.

Designing for agents. The web is shifting from machine-to-human pages to agent-to-agent traffic: Vercel now serves agents precise, structured signal instead of a marketing site, and Rauch argues every business must build agent-native interfaces or be left out of what he calls the most exciting economic opportunity of our times.

SaaS to self-generation. The lethal combination, he says, is buying the platform once and never buying software again: companies will increasingly generate their own tools and agents rather than licence bespoke SaaS, shifting spend toward broad platforms like Vercel.

Code as responsibility. Rauch treats code as a liability — it can carry vulnerabilities or leak customer data — and as code comes to run ever more of the world, from Formula 1 cars to hospital systems, he argues verification and AI safety must be taken seriously, distinguishing critical capital-S systems from playful vibe coding.

Read more: Sourcery (YouTube) · Sourcery (Spotify)