Anthropic’s Claude Code team: manager practice, pod autonomy, and review gates

First-person account: Fiona Fung, director of engineering for Claude Code and Cowork, Running an AI-native engineering org (June 3, 2026). She describes making every new manager start as an individual contributor and ship, so managers understand the changed practice; a small set of non-negotiables includes using the product across functions, a flat team and permission to retire obsolete processes. Pods choose their own triage and planning rituals and what to automate. Roadmap and design-doc work shifted toward just-in-time PR discussion, prototypes and feedback from internal users. She personally automated daily customer-feedback-channel summaries. She says human review remains for domain judgment: legal partners, security/trust-boundary experts, PMs and designers; model handles routine style, lint and some bugs/tests. She offers onboarding ramp, PR cycle time and Claude-assisted commits as gauges while warning throughput is not the ultimate goal. These are the company’s reported practices and impressions, not independent before/after effects or a trial of manager-first onboarding.

Concrete shareable-output affordance, not a measured adoption intervention: Anthropic’s June 18 Claude Code artifacts announcement offers versioned team-shareable pages for PR walkthroughs, incident timelines, release checklists, and manager summaries of merged changes. The post says debugging artifacts were common in its internal tests; it reports no measured behavior change or independent quality effect from sharing those pages. Useful to specify the object in Dru’s ‘visible useful work’ hypothesis: a reproducible issue summary or review packet with links, owner and known uncertainty, not a ceremonial adoption dashboard.

Interpretation. The operator account shows a plausible mechanism: managers gain lived experience, teams can inspect work objects, pods adapt workflow, and risk specialists retain review authority. It does not identify whether manager involvement, extra staffing, model quality or other factors caused faster onboarding. Contrast team-randomized incentives registration and the team-adoption hypothesis; link to cross-functional responsibility and review gates.