Summary
Members of Anthropic’s Claude Code team sit down to discuss how their own software development workflows have evolved over the past year — and how they now rely on Claude Code and a Slack-native agent called Claude Tag for the majority of their daily work. The conversation offers a rare first-party look at how the people building an AI coding assistant actually use it.
Team members describe shifting from carefully reviewing individual tool calls and transcripts to simply setting high-level goals and letting Claude Tag operate autonomously. One engineer notes that 70–80% of their work now flows through Claude Tag, with the remaining 20% handled in the TUI or desktop app for fine-grained control. They discuss how Claude Tag’s access to Slack context — product decisions, architectural rationale, team history — dramatically improves the quality of its choices compared to isolated coding prompts.
The conversation also covers how AI has transformed code review: Claude handles low-level nit-picks autonomously, freeing human reviewers to focus on service boundaries, API design intent, and broader architectural concerns. The team explains how this need to fan out and coalesce review results led directly to the development of Claude Code’s workflow primitives. Throughout, they reflect on the unusually short shelf life of AI tooling — features built to compensate for model limitations become obsolete every few months as underlying models improve — and how that demands a fundamentally different product philosophy.
📺 Source: Claude · Published September 02, 2026
🏷️ Format: Interview
