Running an AI-Native Engineering Org: How Anthropic Rebuilt Its Processes Around Claude Code

The Challenge Nobody Talks About

When agentic coding shifts from an individual productivity tool to an org-wide default, the bottleneck moves. Coding itself stops being the constraint, and everything built around the assumption that coding is slow suddenly becomes the problem. Fiona Fung, Director of Engineering for Claude Code and Claude Cowork, walked through this transition at Code with Claude SF 2026.

Her core observation: "Verification, code review, and security took their place" as the new bottlenecks. The engineering processes that made sense when writing code was the hard part become noise, or worse, friction, when the hard part is now evaluating what was written.

What Changed at Anthropic

Planning

Six-month roadmaps gave way to just-in-time planning with rapid prototyping loops. When implementation time compresses dramatically, the value of long-horizon planning drops: decisions that would take months to validate can now be tested in days. Anthropic's team shifted to shorter planning cycles with higher prototyping velocity.

Context Gathering

The traditional approach (finding the code's author and asking them) was replaced by querying Claude directly. Knowledge that previously lived in people's heads became accessible through the codebase itself, reducing the bottleneck of finding the right person.

Code Review

Review was rebalanced. Claude handles style enforcement, linting, and routine bug patterns. Human reviewers shifted focus to expertise-driven concerns: architectural decisions, security implications, and correctness in domains where automated review is unreliable.

Role Boundaries

Team roles blurred in ways that expanded capacity. PMs began writing code; engineers began handling design work. The constraint was no longer "can this person code": it shifted to judgment, taste, and systems thinking.

Organizational Principles

Fung's recommendations for engineering leaders managing this transition:

• Keep the team flat: every manager starts as an individual contributor to internalize what effective engineering looks like in the new environment.
• Question every process explicitly: team members have explicit permission to challenge and retire processes that no longer serve their original purpose.
• Hire for builders and systems experts rather than raw coding throughput: the value of "can write code fast" has compressed dramatically.
• Track the right metrics: onboarding ramp time, PR cycle time, and percentage of Claude-assisted commits are the leading indicators of the transition's progress.

What This Means for the Field

The talk represents something unusual: a first-person account from the engineering leadership of the team building Claude Code, describing how Claude Code changed their own engineering practice. The feedback loop (building the tool while using it) gives the observations a credibility that external case studies lack.

The full session recording is available on YouTube.