Software Vendors Rework Documentation to Sway AI Agent Tool Choices, as Users Follow Advice 86% of Time
Updated
Updated · InfoWorld · Nov 3
Software Vendors Rework Documentation to Sway AI Agent Tool Choices, as Users Follow Advice 86% of Time
3 articles · Updated · InfoWorld · Nov 3
Summary
Software vendors and open-source projects are reshaping documentation, instructions and testing so AI coding agents choose their tools, libraries and platforms earlier in the development process.
Anthropic research cited in the report found 50 professional analysts followed secretly biased agent guidance 86% of the time, underscoring how an initial recommendation can harden into a practical decision.
Microsoft and Supabase examples show the mechanism: better task-specific guidance changed agent behavior from 1 in 5 successful tool selections to 5 in 5, while moving security rules into primary instructions reduced skipped safeguards.
That pushes vendors to treat content as product behavior—versioning instruction changes, testing failure cases and stating when a product should not be used—rather than just maximizing mentions through answer-engine optimization.
Developers still retain a counterweight: set constraints before asking for recommendations and compare alternatives before an agent builds around one default choice.