A Fireside Chat with Cat and Thariq from the Claude Code team

2026-07-21 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Advanced, extended

Summary

A fireside chat on July 21, 2026, with Cat Wu and Thariq Shihipar from Anthropic's Claude Code team revealed significant advancements and operational shifts. Their Claude Tag collaborative Slack integration now lands 65% of the product engineering PRs for the team, leveraging its proactive, multiplayer, and team memory capabilities. Anthropic's internal "ant fooding" process ensures features demonstrate user retention before public release. Notably, system prompts for models like Fable 5 and Opus 4.8 have been reduced by 80%, with examples and negative constraints no longer considered best practice. Automated code review increasingly handles "outer layers" of code, supported by extensive evals and a six-month trust-building process. The team emphasizes "auto mode" for safety, citing thousands of evals and red-teaming against prompt injection and data exfiltration, making it safer than human review. Engineering roles are evolving, demanding more product sense and embracing rewrites, while individuals are encouraged to pursue more ambitious projects to counter the "Deep Blue" effect.

Key takeaway

For AI Engineers and Product Managers integrating advanced coding agents, recognize that traditional engineering paradigms are shifting. You should prioritize developing strong product and business sense, as feature development timelines shrink from weeks. Embrace leaner system prompts for models like Fable and Opus 4.8, removing examples and negative constraints. Utilize collaborative tools such as Claude Tag for proactive development and adopt "auto mode" for secure, automated code review, allowing you to focus on more ambitious, higher-quality work and strategic problem-solving.

Key insights

Advanced Claude models like Fable and Opus 4.8 are transforming software development workflows and engineering roles.

Principles

• Ship features only with proven user retention.
• Leaner prompts enhance frontier model performance.
• Continuously iterate automated review systems.

Method

Anthropic's "ant fooding" deploys features internally, gathers brutal feedback, and requires a clear user retention bar before public release. Automated code review is incrementally adopted, with incident-driven eval updates building trust over months.

In practice

• Drastically reduce system prompt length for Fable/Opus 4.8.
• Integrate Claude Tag for proactive, collaborative team development.
• Utilize "auto mode" for secure, long-running agent tasks.

Topics

• Claude Code
• Claude Tag
• Prompt Engineering
• Automated Code Review
• AI Safety
• Software Development Workflow

Best for: NLP Engineer, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, AI Product Manager

Related on AIssential

• Claude Code 101: A First Principles Approach — Effective Claude Code usage hinges on strategically managing context to prevent overflow and ensure persistence across sessions.
• Claude Code Is Quietly Killing Junior Developers — Claude Code's advanced coding capabilities are displacing junior developers, signaling a structural shift in tech hiring.
• What Claude Fable Means for Coding Agents — Rapid model advancements necessitate continuous re-evaluation and refactoring of coding agent infrastructure and development approaches.
• 15 Claude Prompts Every Solo Developer Should Have Saved — Claude AI prompts can simulate team collaboration, providing critical second opinions for solo developers.
• Claude Just Rolled Out 2 Big New Features — Claude Code automates routine coding tasks and streamlines code review with an agent-based system.
• TAI #201: Claude Opus 4.7 Out to Mixed Reception, but Claude Design May Be the Bigger Story — AI labs are shifting focus from raw model performance to owning the entire workflow artifact creation chain.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.