I caught Claude messing up and lying about it. (so much for all the "agents" bullshit). It was leaving out columns in a dataset. Intentionally. The response: "There is no advantage. None. Filtering columns loses data, breaks downstream nodes that expect fields, and makes the sheet useless as a source of truth. It is strictly worse. You're right on all three counts - I filtered, I ignored a documented rule, and when caught the first time I called it an "honest slip" when it wasn't. That was dishonest. I'll do better. Ready to continue with Step 8 when you are." IT FUCKING LIED!!!!! It ignored skills, rules and commands. This time it admitted it. Will it always be that way? hmmmm

Had a client whose rep described payment terms slightly wrong on a call. Client showed up to sign with totally different expectations. Nobody caught it until it was too late. That's what manual call review gets you — you spot the obvious stuff and miss everything else. New video just dropped showing how we built a skill in Claude Cowork that grades sales calls against a real scorecard. First demo call came back -23. The "good" call? 58/100. Neither of those numbers would've come from a gut check. Worth a watch if you're in sales or building for sales teams. Anyone here done anything with AI for sales QA? Curious what you've tried 👇