Building the most AI-pilled engineering team in the world | Fiona Fung (Anthropic)

TL;DR

• 1

  Anthropic's code output jumped 8x, so verification became the new bottleneck: Fung says the hard part is no longer producing code but checking quality, impact, and user experience as engineers, PMs, and designers all start committing code.

• 2

  Managers are using Claude as an operating layer, not just a coding tool: Fung runs Claude sessions across repos, Slack feedback, and metrics so monthly reviews become evidence-based conversations about outcomes, bugs, and themes instead of anecdotal status updates.

• 3

  The strongest hires now fall into two buckets: Anthropic looks for creative builders with product sense and deep systems experts who can verify the hard parts, especially in distributed systems and other trust-but-verify domains.

• 4

  High agency is the team norm, but only with high accountability: Fung wants people to take initiative, propose hypotheses, and ship boldly, while staying responsible for what broke, what worked, and whether the work actually mattered.

• 5

  AI can make engineering lonelier, so Anthropic is adding human rituals back in: After noticing people were mostly working solo with agents, the Claude Code team started pairwise programming lunches and hackathons to share workflows and keep team energy up.

• 6

  Fung is most worried about the next generation of engineers: If models abstract away more code over time, she thinks the big open question is how newcomers still learn architecture, dependencies, and deep technical intuition, possibly through more apprenticeship-style paths.