Agents
55d agoSearch Agents Struggle With Query Ambiguity
The DiscoBench benchmark reveals that AI search agents score just 43% accuracy on ambiguous queries. Models often loop through repeated searches rather than asking users for clarification. This behavior yields a 51.9% success rate, which is worse than simple guessing.
The Signal
Removing ambiguity boosts accuracy by 40 points, highlighting a critical gap in agentic communication.