CHAPTERS
Context: Leading Claude Code & Cowork in a rapidly shifting engineering landscape
Fiona Fung introduces her role leading Engineering and Product for Claude Code and Cowork and frames the talk as a set of lessons learned. She outlines five themes: shifting bottlenecks, rewriting norms, rollout approach, proof signals, and a takeaway exercise.
The core shift: Bottlenecks moved from coding to everything around coding
The talk’s central claim is that coding is no longer the limiting factor on AI-native teams; upstream and downstream processes become the new constraints. Fiona emphasizes continuously auditing whether existing norms still serve their original purpose.
A historical analogy: From on-prem build queues to cloud CI—bottlenecks always evolve
Fiona recounts early-2000s build constraints at Microsoft (limited merges, queued builds, debugging failures) to show bottlenecks have shifted before. Today’s AI shift is another iteration of that pattern.
Old bottlenecks fade: TDD and refactoring become easier (and even enjoyable)
With Claude’s assistance, tasks that used to feel costly—like test-first development and large refactors—become much less painful. Fiona shares personal onboarding stories where Claude reduced the “tax” of good engineering practices.
New bottlenecks: Verification, review responsibility, and long-term maintenance
As throughput increases, correctness and quality assurance become the primary constraints. Teams must rethink who reviews changes and how to keep systems maintainable when more people can ship more code faster.
Processes that quietly stopped working: planning, ownership, reviews, team design, knowledge sharing
Fiona lists several traditional processes that need rethinking in AI-native environments. When Claude co-authors most commits and roles blur, classic assumptions about planning rigor, ownership, and documentation need updates.
Technical decisions in an AI-native org: ‘Code wins’ and prototyping as the debate format
Instead of lengthy whiteboarding and speculation, the team resolves debates by building multiple versions and comparing real code. Prototypes become safer and more valuable because iteration and scaling are faster with model help.
Reducing heavy design docs, doubling down on verification and ‘shift-left’ automation
With more implementation capacity, the team reduces long design documents in favor of PR- and prototype-centered discussion. The focus shifts to catching issues earlier through automation and verification closer to the source.
Rethinking ‘who changed this?’: Ask the real question and use Claude to recover context
Rather than relying on ownership heuristics, Fiona encourages teams to identify what they truly need—blame, context, or explanation—and use Claude to help answer it. This reframes debugging and collaboration in co-authored codebases.
Code review in practice: What Claude handles well vs. where humans must stay in the loop
Claude Code Review helps scale reviews by catching style issues, obvious bugs, and spec drift—especially when specs live in the repo. Humans remain essential for risk, legal, trust boundaries, and product taste.
Team makeup for AI-native engineering: creative product builders + deep systems experts
As roles blur, Fiona emphasizes two profiles: builders with strong product sense and engineers with deep system expertise. She also shares how to develop product sense through dogfooding, shipping, and direct customer contact.
Cross-functional gap closing: designers ship polish, engineers get content help—with Claude
Claude enables non-engineers to implement changes (e.g., designers making UX polish updates) and supports engineers in areas like content design. This speeds iteration but increases the need for strong verification practices.
Managers as ICs first: dogfooding culture, faster onboarding, and makers’ time regained
On Claude Code, every manager starts as an individual contributor to learn the codebase and workflow firsthand. Fiona argues AI reduces the friction of returning to hands-on work and makes onboarding less taxing for the rest of the team.
Knowledge sharing resets: the codebase (and checked-in specs) becomes the source of truth
Instead of relying on documentation that can quickly drift, Fiona positions the repository as the authoritative reference. Claude can provide “tech deep dives” on-demand by querying the code and nearby surface area.
Rolling out new norms: top-down principles + bottom-up adaptation, then measuring proof
Fiona describes a rollout model: align on a few non-negotiable principles while letting pods customize workflows. She closes with leading indicators (onboarding time, PR cycle time breakdowns, Claude-assisted commits) and an exercise to ‘Claudify’ the noisiest workflow.
