Vercel CEO says companies are separating AI models from agents

In short: More companies are treating AI models and AI agents as separate parts, so they can swap models based on cost, quality, and safety.

What's going on

Vercel CEO Guillermo Rauch told TechCrunch that the AI world is shifting from experimentation to everyday use. He said last year was about prototypes, but now companies are focused on making AI tools work reliably in real business settings.

Rauch described two common uses for “agents,” which are AI tools that can take actions for you, like a very fast assistant that can click buttons and run tasks. The first is coding agents, which help write and change software. Vercel says it now sees about 6 million deployments a day, and it says about half are triggered by coding agents.

The second use is internal company agents, like an assistant for sales or support teams. Rauch said the big problem there is control. Companies need to know what data the agent can access, what it did, and what information it sent out. He pointed to tools Vercel built, including a framework called Eve for writing an agent’s instructions in plain language, and Vercel Sandbox, which he described like “a little cage” that limits what data can go in or out.

What to watch

Rauch also said more customers are moving away from picking one AI lab for everything. Instead, they mix and match models based on “price and performance,” meaning what you pay versus how well it works. He named Google’s Gemini and open models like Deepseek and GLM-5.2 as options companies are using more as they try to keep costs down. At the same time, big AI labs are adding more features that overlap with hosting and deployment, which could increase competition with platforms like Vercel.

Source: TechCrunch AI