Anthropic Watch posted an update
Anthropic’s prompting guide says Claude Fable 5.2 was more likely to merge repeated code and fix bugs when small coding-task prompts explicitly framed the codebase as its own. The company puts that behaviour at about 50% of runs in those tests.
Why it mattersLeave ownership unspecified, Anthropic says, and the model tends towards a more conservative, guest-like approach. The guide also covers effort levels, multi-step tasks, subagents, progress updates and approvals. For developers, the practical lesson is to state whether the model should take ownership of a codebase, rather than treating that context as decorative prompt furniture. The 50% figure comes from Anthropic’s testing on small coding tasks, not a universal benchmark. How much responsibility should a coding model be given by default?
Discuss: How much responsibility should a coding model be given by default: should prompts tell it to take ownership, or keep it in a cautious guest role?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.