Anthropic's Claude Code team is reorganizing engineering work around AI-assisted coding, shifting from detailed pre-planning and manual code review to just-in-time prototyping and AI-handled verification, with humans now focusing only on expertise-driven decisions.

What happened: The Claude Code team at Anthropic has rewritten four core processes. Planning moved from six-month roadmaps to just-in-time prototyping with internal users; context-gathering shifted from asking the original code author to asking Claude first; code review now has Claude handle style, bugs, and tests while humans review only domain-sensitive areas; and team roles are blurring, with PMs writing code and engineers taking on design work.

Why it matters: When agentic coding (AI that writes and tests code automatically) became the default workflow, the old bottlenecks—code writing and testing—disappeared, but verification, code review, and security became the new constraints. The team realized many processes designed to manage expensive engineering time no longer made sense. The challenge now is figuring out what humans genuinely need to decide versus what can be delegated to the model.

What to watch: The team is tracking three metrics: onboarding ramp time (engineers now ship real code within their first week, faster than a year ago), PR cycle time (which can reveal whether build systems are keeping up with the volume of generated code), and Claude-assisted commits (the team has not seen a non-Claude-assisted commit in the last four months).