DoorDash has welcomed the team behind Metis into its ranks, announcing the acqui-hire as part of a new DoorDash AI Research division. Co-founder and CTO Andy Fang made the announcement on X, saying the company had been partnering with Metis for six months before deciding to bring the team in-house.

Metis, a Y Combinator Summer 2025 startup, built what it described as the "post-training and continual-learning layer for enterprise agents." The 13-person San Francisco team — led by founders Aryan Shah, Aayush Sheth, and Marcus Yearwood — developed infrastructure to make AI agents more reliable in production by training on company-specific data, running reinforcement learning loops, and evaluating agents against real-world tasks before deployment.

Why DoorDash wants an agent reliability lab

DoorDash has been steadily building AI into its product surface. Earlier this month, Fang highlighted an AI-powered upgrade to Zesty, DoorDash's agentic restaurant recommendation app, as a preview of what agentic capabilities are coming to the main DoorDash ecosystem.

Absorbing Metis gives DoorDash a dedicated research team focused on making those agents trustworthy at scale — critical for automating anything from order handling to logistics routing in local commerce.

Fang framed the ultimate goal as two-pronged: agentic commerce, where AI agents transact on behalf of users, and physical intelligence, where autonomous systems extend into the real world of deliveries and logistics.

"It's still early innings with how AI will transform local commerce," Fang wrote, promising more details soon.