Building the most AI-pilled engineering team in the world | Fiona Fung (Manager of the Claude Code and Cowork Teams)
with Fiona Fung
21 Jun 20264 min read1h 05m
TL;DR
Anthropic engineers ship 8x as much code per quarter compared to 2021–2025, and Fiona Fung argues this makes coding no longer the bottleneck — the new constraints are ambition, verification, and quality oversight. She describes how she uses Claude Code sessions and automated routines to stay on top of an explosion of output across her teams. The engineers thriving most are those with a growth mindset who treat AI as a ceiling-raiser rather than a threat.
Key Moments
Fiona Fung
“We say with high agency is also high accountability. So, it's all about making sure folks have that freedom to code. But, then it's also like, okay, what's the accountability for it? What's the hypothesis of what you're trying to solve?”
Fiona describing the cultural principle her team uses to balance speed and ownership when everyone has AI-amplified coding ability.
“Coding is no longer the bottleneck. And so now like, you know, you showed up you showed the the tweet and that graphic. So but now it's all about like where has that shift happened? Like now not only engineers but we also have designers, PMs, everybody on the Clockwork team checks in code.”
Fiona explaining the broader implication of the 8x productivity stat — it's not just engineers shipping more, it's everyone shipping.
“I actually have a cloud code remote session that I enlist in all of our repos. And so, this way I have full visibility into the work that everybody's doing. And this instance it also has access to all our Slack channels and and I'll I'll have access to like how how are the metrics of of everything we track.”
Fiona describing her new management practice of running a persistent Claude Code session across all repos to stay on top of what her team is shipping.
“I think if you have specs or like or like like check those into the repo and then make sure the spec also keeps up to date with the code like frequently. But that's what I found like works really well of any time you have like a statement of what good looks like, get that into the repo and then Claude code review can make sure it's still matching what you set up to do.”
Fiona explaining how she uses spec-driven development as a framework for automated Claude code reviews to maintain quality at high velocity.
Fiona Fung leads the teams behind Claude Code and Co-work at Anthropic, overseeing engineering managers Boris Cherney and Kat Wu. Before Anthropic, she spent 11 years at Microsoft building TypeScript and Visual Studio, then moved to Meta where she launched Facebook Marketplace (now $100B+ GMV/year), worked on Meta's first smart glasses and AR product Orion, and led Instagram infrastructure, growth, integrity, and safety teams overseeing 500+ people. She has been a software engineer for over 25 years.
Takeaways
1
Persistent Claude session replaces manual management check-ins Fiona runs a Claude Code remote session enlisted across all repos, with access to Slack channels and team metrics, using it monthly to review what shipped and how it performed. This replaces the old bullet-list status update and allows a manager to have substantive, data-backed conversations with direct reports at a scale previously impossible.
2
Specs in the repo enable automated quality gates Checking written specifications — what 'good looks like' — directly into the codebase allows Claude code review to validate every PR against those standards automatically. This is described as the evolution of test-driven development: instead of writing tests first, you write the definition of quality first and let the model enforce it.
3
Automated morning routines replace manual feedback triage Fiona replaced her daily ritual of manually scanning feedback channels with a Claude routine that aggregates themes overnight and surfaces ready-to-review PRs by morning. This compounds the productivity gain: not only is code generated faster, but the management layer that routes and prioritizes fixes is also automated.
4
8x output shifts bottleneck to verification Anthropic engineers ship 8x as much code per quarter compared to 2021–2025. When coding ceases to be the constraint, the new bottleneck becomes verifying that what ships is correct and impactful — requiring new processes around specs, automated review, and quality frameworks rather than more human reviewers.
5
Hire creative builders and deep systems experts — not generalists Fiona found that Claude Code's early team of product generalists was missing systems and distributed systems expertise, which became a gap as scale demands grew. Her hiring framework now targets two profiles: product-sense dreamers who own features end-to-end, and deep subject matter experts for the hard parts models still can't reliably verify.
6
Pair programming lunches counter AI-driven isolation The Claude Code team noticed engineers were becoming lonely from working almost exclusively with their AI agents, so they started pair-wise programming lunches to restore human collaboration. This is an early signal that high AI adoption requires deliberate social infrastructure to prevent team cohesion from eroding.
7
AI raises the ceiling for non-experts in any domain A non-mobile engineer at Anthropic extended a feature to Android surfaces with Claude as a partner, something he would have previously declined as out of scope. The implication is that cross-functional scope expands dramatically — engineers, PMs, and designers can all ship code — which changes how teams are structured and how work is assigned.