Vercel's agent story is moving quickly, but the verified timeline matters: Vercel said agents were near 29% of deployments in mid-June, and TechCrunch reported they had reached half by early July.
Vercel is no longer just the place where developers push web apps. According to TechCrunch, the company now sees 6 million deployments a day, with coding agents triggering about half of them, and more than 1 trillion tokens moving through its AI gateway daily. That is the shift. If you build software for a living, you can already feel where this is going: the deployment target is becoming part of the agent workflow, not the last step after a human finishes coding.
The original number needs tightening. Vercel's own eve launch post on June 17 said agents had triggered less than 3% of deployments a year earlier and were around 29% at the time, with the company expecting half of deployments to come from agents soon. By July 6, TechCrunch reported that half of Vercel's daily deployments were being triggered by coding agents. Those two facts tell the story more cleanly than forcing the 50% figure into the London keynote itself.
It happened fast.
Rauch's argument is not that every developer disappears behind a prompt box. It is that agents are becoming first-class software builders, and they need somewhere to run the code they generate. TechCrunch also reported in April that 30% of the apps running on Vercel's platform already came from agents, while Vercel's annual recurring revenue had risen from $100 million at the start of 2024, as reported by The Information, to a $340 million run rate by the end of February 2026, according to Forbes. You don't need to dress that up. The infrastructure business got a new buyer, and sometimes that buyer is software.
Vercel is selling the agent layer
At Ship 2026, Vercel put product names under the pitch. Its London announcements included eve, an open-source framework Vercel described on stage as a kind of Next.js for agents. In practice, an eve agent lives in a directory: durable execution, sandboxed compute, approvals, subagents and evals, all wired in. That matters. Production agents are not useful if they can only write demo code on a laptop - they need an identity, a log trail, hard boundaries, and a way for a person to stop a bad action before it hits a real system.
Vercel Connect is aimed at the same problem from the security side. Vercel says it replaces long-lived stored tokens with scoped credentials generated at runtime, so an agent can access GitHub, Slack, Notion, Salesforce or another service without carrying a permanent secret around in an environment variable. A vault can hide a dangerous token. It can't make the token less dangerous once it leaks. Vercel is right to make that distinction.
Vercel Services adds another piece: backends, frontends and other services can be deployed inside one coordinated project, with private communication between components that does not touch the public internet. For a normal web team, that is useful plumbing. For an agent that is writing, testing and shipping software with less human hand-holding, it becomes closer to guardrails.
Rauch's bigger strategic line, laid out in his July TechCrunch interview, is about separation. The labs want to pull more of the stack into their own products. OpenAI, Anthropic and Google have every reason to make creation and hosting feel like one surface - execution included. Rauch is arguing for the opposite: models, agents, sandboxes, gateways and deployment targets should remain pieces you can swap. That is not a neutral opinion. It is Vercel's commercial position, and it is also the pitch many developers will prefer if they don't want one lab deciding the whole shape of their software stack.
The rivals have moved too
The competitive section also needed a factual correction. It is no longer accurate to say Netlify and Cloudflare have nothing comparable in the agent direction. Netlify's current docs describe Agent Runners that work with Claude Code, OpenAI Codex and Google Gemini, generating deploy previews for agent-made changes. Cloudflare announced its Agent Cloud in April and its Agents SDK docs were still being updated in July. The market saw the same turn Vercel saw.
Vercel still has a strong claim here because deployment is already central to its identity. Next.js and v0 give it a natural path from prompt to running application - the hosting workflow follows. Cloudflare has the edge network and Workers. AWS has enterprise gravity. Netlify has moved directly at agent deployment. So the fight is not Vercel against sleeping incumbents. Frankly, that would be too easy. The fight is whether Vercel can make agents choose its platform by default while everyone else teaches agents to do the same thing somewhere else.
The IPO thread fits inside that. TechCrunch reported in April that Rauch signaled public-market readiness at the HumanX conference in San Francisco, while declining to give a specific quarter. Vercel was last valued at $9.3 billion after a $300 million Series F co-led by Accel and GIC in September 2025, according to Vercel's own announcement and Reuters coverage. That valuation came before the latest public attention around agent-triggered deployments, but not before investors understood the AI angle.
The hard question is not whether Vercel has found a good narrative. It has. The hard question is whether the agent deployment curve stays steep once the big model companies, cloud providers and hosting platforms all optimise for the same workflow. So far, the numbers say Vercel has real momentum. You should still watch the source of that momentum carefully: agents are prolific, but defaults can change quickly when the companies building the models also control the front door.
Also read: Bankr bot's return to X after suspension exposes how fragile crypto's social layer really is • A Chinese AI model just turned a routine training trick into a Washington national-security crisis • Two-thirds of all venture capital is now flowing to AI startups and non-AI founders are feeling it
