AI Agents Speed Up Code Writing, But Merging Pull Requests Now Slows Teams

A developer reports that AI agents can produce around twenty pull requests in an hour, yet merging them consumes the rest of the day. Each merge tends to create conflicts for later ones, and continuous integration checks add further waiting, especially for feature changes that touch many files. Anthropic reportedly saw merges into its monorepo rise from about 2,500 a day in May to more than 30,000 in September, prompting it to build its own source control system, Scion. The author argues that the bottleneck has shifted from writing code to managing it after it is written. Existing workflows were designed for humans opening a few pull requests daily, and stacked pull requests offer only partial relief.