AIAgent memoryStory 03
Simple memory beats elaborate agent systems
What changedResearchers compared a plain memory recipe with systems that rewrite and reorganize what an agent remembers. Dense retrieval over the raw conversation ranked first in six of seven benchmark-and-model settings, beating the strongest agentic alternative by 0.13 to 5.53 points.
The useful part
Why it matters
Keeping the original conversation may preserve useful links that get lost when an agent repeatedly summarizes its past.
Keep in mind
Good to know
The result covers four conversation-memory benchmarks, not every long-running agent task.
Evidence