One category that seems genuinely useful is agents that help reason about complex systems rather than directly execute tasks.
For example, I’ve been experimenting with a small tool that takes a description of an AI product and decomposes it into tasks, estimates token usage, and models the cost implications of different architectures.
It’s not an autonomous agent, but it’s been surprisingly useful as a “thinking tool” before building something.
For example, I’ve been experimenting with a small tool that takes a description of an AI product and decomposes it into tasks, estimates token usage, and models the cost implications of different architectures.
It’s not an autonomous agent, but it’s been surprisingly useful as a “thinking tool” before building something.