Descriptions:
Fiona Fung, who leads Claude Code and Cowie engineering and product at Anthropic, shares hard-won lessons from building and managing an AI-native engineering organization. Drawing on experience across Meta, Microsoft, and Anthropic, she outlines five themes she’s observed as Claude Code fundamentally shifts where bottlenecks live in the software development process — and why practices that served teams well in the past may actively work against them now.
The central argument is that coding throughput is no longer the expensive constraint. On the Claude Code team, code generation is rarely the slow part; verification, decision-making, and context management have taken its place. This inversion forces teams to rewire norms around planning, code review, and quality assurance. Fung describes adopting a “shift left” philosophy — catching issues earlier through automation — and explains how Claude Code’s own review tooling now plays a central role in the team’s PR workflow. She also addresses the blurring of traditional roles: designers submitting code, and the question “who made this change?” becoming harder to answer and less useful than the underlying questions it was really trying to ask.
The talk includes candid reflections on Anthropic’s own Routines feature, which Fung notes wasn’t even in her slide deck when she started writing the talk, illustrating how fast the tooling is evolving beneath the teams building it. She closes with open questions she’s still working through and practical prompts for engineering leaders looking to start similar conversations with their own teams.
📺 Source: Claude · Published May 08, 2026
🏷️ Format: Opinion Editorial
