Anthropic's engineering team published a post on October 13 called "Creating in-house source control." The striking detail is how quickly the company's review workload changed: it says it merged around 2,500 code changes per day in May, and more than 30,000 per day in September.
At that pace, Anthropic says it became "bottlenecked by developer infrastructure built for human (rather than agent) scale." Its response was to build a new version-control system and migrate its entire monorepo off Git in five months.
Why this is more than an internal IT story
The episode is a concrete example of a wider shift. As AI agents write more code, the constraint moves from writing to reviewing, integrating, and shipping. Tools and habits designed around a human pace of commits—review queues, merge conflicts, slow CI—can quietly become the limiting factor, even when the code itself arrives faster.
It also cuts against a common assumption about AI and work. The hard part is rarely generating output; it is absorbing it without losing quality control. Anthropic's answer was to rebuild its own tooling rather than wait for it.
Alongside the engineering post
On October 14, Anthropic also said it opened an office in Singapore—its fifth in the Asia-Pacific region—and signed a memorandum of understanding with the Singapore government on AI safety and responsible adoption. The company has been publishing a steady stream of safety and governance updates, including giving a team from the evaluation group METR access to assess its practices.
Sources: Anthropic's post on source control and its Singapore announcement.
