search precision

Search precision is the ratio of genuinely needed items among everything an AI actually pulls up when asked to find something. Think of looking for a document in a drawer: if there are only 5 documents in the drawer, you can grab the right one quickly, but once there are 100, you're much more likely to grab the wrong one with a similar-looking title.

AI agents run into the same problem. When there are only a few stored task instructions or reference materials, they pick out exactly what's needed. But as more materials pile up, ones with similar names or descriptions get mixed together, and the AI ends up pulling out the wrong material more often. This is when we say search precision has dropped.

So simply stockpiling as much material as possible isn't always the best strategy. No matter how good a set of instructions is, it's useless if it can't be retrieved accurately at the moment it's needed. That's why, these days, how accurately something can be found matters more than how much is stored.

Articles use it like this: "When the skill library had 5 items, actual search precision was 29.6%, but once it grew to 100 items, it dropped to 3.3%." This means that as the amount of stored material (skills) increases, the rate at which the AI accurately finds what it needs drops sharply. A common misconception is that piling up more material makes an AI smarter, but this metric shows that as material grows, the risk of pulling up the wrong thing can actually increase.