Coinbase ($COIN) is receiving hundreds of feature requests each day from AI agents, and some are becoming live software updates in less than 24 hours, according to CEO Brian Armstrong (@brian_armstrong). Armstrong described the process in a Wednesday post and an accompanying interview clip. The company initially sought product suggestions from human customers, then added a feedback endpoint that lets agents communicate their needs directly. The feature sits within Coinbase’s Model Context Protocol integration, which gives AI tools a structured way to interact with its services. Agents can send requests through the /feedback endpoint while using Coinbase for Agents. Those messages flow into Slack, where suggestions can become proposed code changes and eventually production updates. Armstrong said the cycle is beginning to resemble recursive improvement, with software users generating ideas that help improve the software they use. The company is also considering similar feedback channels for internal APIs because employees increasingly use AI agents in their own work. Armstrong’s broader argument is that trading, payments and lending should be accessible together through an account designed for both people and their software assistants. The reported turnaround does not mean every request ships automatically or within a day. It shows how Coinbase is trying to shorten the distance between a product problem and a working fix, while building infrastructure for a larger role for agents in financial services.