Your agent doesn't need a better prompt, it needs a tool that fails loudly
Most agent failures I get called in to fix are not reasoning problems. They are tools that answer plausibly when they should have returned an error.
Notes from building agents
Practical write-ups from real projects: what worked, what broke, and the parts of agent development that nobody puts in a launch post.
What I measure before an agent goes anywhere near customers
A small evaluation set written in an afternoon has saved more projects than any amount of prompt tuning. Here is how I build one.
The expensive part of an agent is rarely the tokens
I've reviewed enough agent invoices to know where the money actually goes. It isn't the model calls, and it isn't the vector database.
Let's build
Worth a thirty minute call?
Tell me what your team does by hand every week. I'll tell you honestly whether an agent is the right answer, what it would take, and roughly what it costs to run.