Fiona Fung

Sources#

• Running an AI-native engineering org

Summary#

Leads engineering and product for Claude Code and Cowork at Anthropic; previously built and led teams at Meta and Microsoft (she was on Visual Studio in the early 2000s). Her May 2026 talk "Running an AI-Native Engineering Org" is the org-leadership companion to Cat Wu's product account and Boris Cherny's IC account: she describes how the team's processes were rewritten once agentic coding became the org-wide default. Her through-line: "what served you prior may no longer" — audit every process for whether it still serves its purpose, because the bottleneck has moved.

Note: both Fiona Fung and Cat Wu are described as leading product for Claude Code + Cowork. They are complementary senior leaders — Fung frames herself around the engineering org (managers-as-ICs, review, hiring, norms); Wu is Head of Product (cadence, evals, vision-translation with Boris). Treat the title overlap as shared scope, not contradiction.

The five things that changed#

Her talk is structured around five org shifts once coding stopped being the bottleneck:

1. Bottlenecks moved — from engineering bandwidth to verification, review, and maintenance (Verification as the New Bottleneck).
2. Norms rewrote themselves — TDD lost its tax; refactoring stopped being a budget fight; design docs shrank (Building Is Cheap, Arguing Is Expensive).
3. Source of truth moved — from docs to code/skills checked into the repo (Code as Source of Truth).
4. Org shape flattened — every Claude Code manager starts as an IC (Managers as ICs).
5. Product sense became the scarce skill — built through relentless dogfooding (Dogfooding as Product Discipline).

The historical analogy she leans on#

She frames this as not-the-first bottleneck shift: at Microsoft in the early 2000s, builds were the bottleneck — one server room, on-call weeks, a merge queue of "six PRs at a time," debugging which PR broke the build. Cloud + continuous build dissolved that bottleneck. Agentic coding is "just another shift" of the same kind. The discipline is permanent: processes "quietly stop working," and people don't delete processes on their own — they pile up (her SLA-stacking war story). Hence "explicit permission to kill old processes" as a team principle.

Team principles ("must-dos") + bottoms-up#

The forcing functions she aligns the team on: every teammate uses Claude Code and Cowork; "Claudify everything we can" (if Claude can do it, Claude should); explicit permission to kill old processes. Everything below that — how Claude shows up in triage, standups, which workflows get Claudified first — is left bottoms-up per pod. Balance: align on culture, leave room to adapt.

Key quotes#

• "What served you prior may no longer."
• "Building is cheap, argument is expensive."
• "[TDD] kind of took the tax out… it was actually so much more fun." (writing the failing test first, with Claude)
• "Dog food, dog food, dog food — or as we call it at Anthropic, ant food."
• "Our code is our source of truth."

Connections#

• Claude Code — leads engineering + product
• Cowork — leads engineering + product (sibling product)
• Anthropic — employer
• Cat Wu — complementary product leader on the same products
• Boris Cherny — debated refactoring approaches with her by generating three PRs instead of whiteboarding
• Verification as the New Bottleneck — her central thesis
• Code as Source of Truth — her onboarding/knowledge-sharing rewrite
• Managers as ICs — her "spicy" org change
• Dogfooding as Product Discipline — her product-sense engine ("ant food")
• Harness Shrinkage as Models Improve — "Claudify everything / kill old processes" is the org-process face of harness shrinkage

Sources#

• Running an AI-native engineering org — "Running an AI-Native Engineering Org" talk, May 2026