Anthropic's Fiona Fung on running an AI-native engineering org

Fiona Fung, who runs engineering for Claude Code and Claude Cowork, writes about how her team works now that agentic coding is the default. Six-month roadmaps are gone in favor of what she calls just-in-time planning, with prototypes and internal users in place of thick design docs. Every commit is Claude-assisted by default, and engineers ship real production code in their first week rather than after a month of onboarding.

The more interesting argument is about where work piles up. Code generation stopped being the bottleneck for her team. Verification, code review, and security took its place. Claude handles linting, style, bug-catching, and adding tests, while humans concentrate on the parts that need expertise: security-sensitive paths, legal review, and product calls. Hiring follows: Fung says they look for creative builders with product sense and engineers with deep systems knowledge, not raw throughput. Product managers code, and engineers do design and content work.

The piece reads as thought leadership, not a postmortem. Fung does not show before-and-after numbers and glosses over what breaks when CI and build systems become the new ceiling. Still, the shape she describes matches what other teams are quietly reporting, and the specifics are concrete enough to argue with: kill obsolete process, dogfood your own tools.

Why it matters

If you lead an engineering team, the bottlenecks she names are coming for your org if they have not arrived yet. Decide now where humans stay in the loop and where you let the model run, and start watching your CI and build infrastructure before they become the wall. If you are an IC, the kind of work that gets you hired is shifting toward judgment and systems thinking, not lines of code per day.