Anthropic says 'double-check your work' prompts are now an anti-pattern — one dev found 125 in his config

Frequent-Ad-836 · reddit · 2026-09-10

Anthropic published a post on reducing cost and improving performance, with a striking core message: the nudging instructions we added for older models now hurt more than help.

Key points

Old models only did what you told them, so phrases like "double-check your work" and "be maximally thorough" were necessary. Current models already go the extra mile, so these prompts make them work twice — costing more and yielding worse answers.
Concrete examples from the post: "be maximally thorough" triggered dozens of unnecessary knowledge-base searches; "double-check your work" caused rework of finished output; contradictory rules (allowed in one place, forbidden in another) meant four valid refunds never went out.

Important nuance

This isn't an argument for stripping everything. Remove the nudges, keep hard constraints (e.g., "never click publish twice"), and use the dedicated effort parameter instead: too low answers prematurely, too high overthinks easy tasks.

The author audited his own rule files afterward: 66 instances of "must" and 54 of "never" — 125 total — and suspects most are deletable nudges rather than real stops.

Original post →

Anthropic says 'double-check your work' prompts are now an anti-pattern — one dev found 125 in his config

More from coding & agent