DoorDash says its internal generative-AI platform has grown to more than 5,000 users, with about 45 new users joining each day. Forty percent are now outside engineering, including people in legal, sales, operations and strategy.
The team began in 2023 expecting to support machine-learning specialists, then found that ordinary product engineers wanted APIs rather than notebooks and GPU access. It built a common gateway and SDK so teams could switch among OpenAI, Anthropic, Gemini and open-weight models without rebuilding provider-specific plumbing.
DoorDash organized the platform around three trade-offs: accuracy, latency and cost. It also chose to support end-to-end product workflows rather than offer infrastructure pieces that each team would need to assemble. As adoption widened, centralized controls and accountability became as important as development speed.
The figures come from a DoorDash conference presentation rather than an independently audited report, and the talk does not reduce platform value to one performance metric. Its practical lesson is architectural: a shared interface makes experimentation faster, but the same gateway becomes the natural place to enforce access, observe usage and compare models under consistent conditions. Broad adoption changes an AI platform from a specialist tool into company infrastructure.