Tool search just turned agent frameworks into library management systems
AI agents are drowning in their own capabilities, and we're solving it like it's 1995.
We’ve built agents with access to thousands of tools, then watched them faceplant because they can’t find the right screwdriver. Tool search is the latest bandage on a problem we created by giving agents more capabilities than they can reasonably track.
The context explosion problem
Modern agent frameworks suffer from success. We can plug in hundreds of APIs, databases, and custom functions. But dumping every available tool into context is like handing someone a phonebook and asking them to find their mate’s number. The BM25 approach that’s showing 49% to 74% accuracy gains isn’t revolutionary search tech. It’s just admitting that agents need the same basic information retrieval we’ve used for decades.
We’re building digital hoarders
The real issue isn’t search. It’s that we’re treating agent capabilities like a collection. More tools equals better agent, right? Wrong. We’re creating digital pack rats that know about everything but can’t decide what to use when. Progressive schema disclosure sounds fancy, but it’s just hiding the mess instead of cleaning it up.
The future isn’t better tool search. It’s agents that know which three tools they actually need for any given task.