The Bottom Line: Claude Code reduces manual oversight through improved model capabilities, while Claude Tag already generates 65% of the team’s product PRs.
Anthropic engineers Cat Wu and Thariq Shihipar report in a technical discussion how Claude Code and the new Fable model have fundamentally changed their daily work — from tedious monitoring of autonomous agents to focused work on higher-value tasks.
In a technical discussion at the AI Engineer World’s Fair, Cat Wu and Thariq Shihipar from the Claude Code team discussed, among other things, changes in their development workflows since Claude Code’s launch in February 2025. During the original launches with Claude Sonnet 3.7 and later Opus 4, autonomous code changes still required intensive manual control — every permission was reviewed and often rejected. With newer model generations, this has fundamentally shifted.
Wu reports that she and her team can now delegate significantly more trust-based tasks to Claude Code instead of monitoring every step. This creates space for design questions: What user experience should be provided when Claude takes on the implementation? With the new Fable model, the value proposition shifts again — many features can now be prototyped in a single step. Shihipar adds that Auto-Mode is now so reliable that manual reviews are barely necessary anymore; he has even used Fable for video editing work that meets the brand team’s high quality standards.
Technically, the teams rely on tiered reviews: critical changes to the Claude Code core are still reviewed manually, while automated code review becomes the standard for the product’s outer layers. In prompt engineering, a paradigm shift is evident: with Fable and Opus 4.8, extensive examples in the system prompt are no longer best practice. The Claude Code system prompt has been reduced by 80%, just as lists of negative instructions (“don’t do X”) can degrade the quality of newer models.
Internally, Anthropic uses a culture of “dogfooding” — internal usage is referred to as “ant fooding”. Claude Tag, Anthropic’s new Slack integration, already lands 65% of the team’s product engineering PRs. The system is being tested internally in public Slack channels, which Anthropic views as a central success factor for product development.
Source: simonwillison.net · Published 21 July 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.