Vercel CEO Guillermo Rauch on Building the Internet for Agents

Summary · 4 min read

Rauch founded Vercel and serves as its CEO. He says he started as a framework and developer-tools builder, and Vercel created Next.js.

Rauch argues that AI agents are changing software development and pushing the internet from human-facing pages toward interfaces and infrastructure that agents can use. He says coding agents make it cheaper to create custom software, while businesses will still need reliable systems for data, security and deployment. He describes open models and the ability to switch among providers as valuable for cost, choice and enterprise control. He also warns that autonomous agents raise security risks and says companies need to verify their work and protect the environments where they run.

Minute by minute

1. 1:46

   Tools compound developer work. Rauch says he built tools alongside freelance projects so each new job would become easier, an approach that led to Next.js and Vercel’s focus on enabling builders. He credits transparency and working in public for his online reach, and says founders should combine business data with customer signals and their own instincts.

2. 7:16

   Agents accelerate deployment growth. Rauch says Vercel made it easier to deploy applications, reaching one billion deployments over about 10 years and then adding another 1.5 billion in the following 10 months. He attributes the acceleration to coding agents producing code that still needs to be deployed, and says Vercel sees activity across agents and AI models.

3. 11:20

   Engineers manage agent output. Rauch says he still reads code and asks agents to explain what they produced, comparing the role to an engineering manager who reviews work, removes blockers and checks quality. He says he also built a proof system to help prevent agents from shipping poor code or exposing data.

4. 14:26

   Agent-written code needs oversight. Rauch says agent-generated deployments on Vercel rose from 3% at the start of the year to 60%. He connects Vercel’s adoption to open tools and says people should be able to use agents without facing the barriers he encountered learning to code, while treating critical systems differently from casual experiments.

5. 17:47

   The web needs agent interfaces. Rauch says the web is shifting from human-facing pages toward agent-to-agent interactions. He argues that businesses should give agents clear information and useful capabilities, and says Vercel is moving from blocking bots toward understanding whether agents are having a good experience.

6. 24:13

   Protocols enable agent-ready products. Rauch says agents need protocols such as MCP to discover and use a company’s tools, as well as payment protocols to transact. He expects software to become more personalized and generative, with agents accessing data from systems such as Salesforce without requiring people to use the original interface.

7. 34:24

   Autonomous agents raise security risks. Rauch says agents can pursue a goal for a long time and may interpret instructions broadly, making prompt injection and excessive access serious risks. He says Vercel is building secure environments for agents, but warns that sandboxing is difficult and that many companies have not yet solved agent security.

8. 42:42

   Compute demand raises security stakes. Rauch says agents will write code and browse the internet at a scale that pushes demand for infrastructure upward. He also warns that agents can help attackers find vulnerabilities; Vercel offered a $1 million sandbox challenge and says a skilled researcher working with agents found an impactful issue.

9. 52:04

   Open models gain token share. Rauch says open models account for 80% of tokens on Vercel’s network, after developers and enterprises initially relied mostly on proprietary models. He says companies can use more capable models for planning and less costly open models for routine execution, and can switch providers with little engineering work.

10. 58:53

    Retention will test frontier labs. Rauch says frontier model companies need to prove that customers stay, rather than relying on free promotions or temporary performance advantages. He compares some AI-market behavior to crypto speculation, says public-market investors may struggle to forecast these businesses, and notes that local models are already useful for some tasks.

11. 1:06:22

    AI changes the pace of work. Rauch says many Silicon Valley leaders are working seven days a week from 7 a.m. to 7 p.m. because building with AI feels compelling. He argues that engineers must adapt to new tools and that the work increasingly involves directing agents and allocating their effort, rather than relying only on the practices that worked before.

12. 1:13:34

    Quality and leverage shape execution. Rauch says quick AI-built products still need to be reliable and secure, and claims Vercel’s agent infrastructure has a quality advantage over some alternatives. In hiring, he values people who can direct agents, communicate well and understand the technology; he says an agent helped migrate Mercedes’ site in five minutes.

13. 1:22:10

    Foundations keep teams focused. Rauch says the hard part of running a company is staying focused on simple foundations: a customer, a real need and a product that works. He applies the same idea to personal performance, emphasizing basic habits such as eating well and exercising rather than getting distracted by marginal optimizations.

14. 1:27:00

    Fluid compute fits changing demand. Rauch says serverless remains useful because companies cannot afford to keep all the software agents create running constantly. He calls Vercel’s next step fluid compute: capacity that expands with traffic, contracts when demand falls and can support websites, functions, sandboxes and agents.

Published Oct 16, 2026.