According to CoinMeta, Coinbase CEO Brian Armstrong stated that Coinbase receives hundreds of functional requests per day from AI agents, some of which are turned into real-time software updates in less than 24 hours. In a post on Wednesday, Armstrong mentioned that Coinbase initially sought product suggestions from human customers, and later added a feedback endpoint to allow agents to communicate their needs directly. This feature is integrated into the model context protocol of Coinbase, providing a structured way of interaction for AI tools. Agents send requests through the /feedback endpoint, and these messages flow into Slack; suggestions can then be translated into code changes and production updates. Armstrong pointed out that this cycle begins similarly to a recursive improvement process, where ideas generated by software users help to improve the software they use. The company is also considering providing a similar feedback channel for its internal API to adapt to the growing trend of employees using AI agents.