A Bank Built Its Own Coding Agent. It Took Off — Felipe Felisola Caso & Francesco Galletta, Monzo

Monzo's coding agent took off fast. Felipe Felisola Caso and Francesco Galletta say it was quick because much of what it needed was already in place. Their talk is summarized here from its YouTube description, not a transcript. The agent began as an emoji reaction in Slack. The speakers say it now authors more than 10% of all merged pull requests across Monzo's tech teams and runs more than 1,800 tasks a day.

Their explanation is lots of little multipliers, meaning capabilities Monzo already had. The first is an MCP gateway into Monzo's internal tools and knowledge. MCP, the Model Context Protocol, is a standard way to connect a model to tools and data. The second is an agent orchestration system that runs work in a sandbox and was built for a regulated bank. The third is a platform that turns the agent itself into a building block.

According to the description, that building block now handles incidents, on-call data quality work, and requests from people who are not engineers, so the agent's use has spread beyond writing code. The talk closes on why Monzo keeps building the agent in-house, though the description does not give the reasons. The same two speakers gave another AI Engineer talk, Stop Chatting With Spreadsheets, about other AI work at Monzo.