You killed the chatbots. Good. The agents you replaced it with are the new problem.

They do real work now. They also make things up, lose the thread, and do the wrong thing with total confidence, and they look great doing all of it. That confidence is why most agent projects stall before production. Nobody can promise the thing will not lie to a customer.

I'm not a developer. I run sales and marketing motions at ArchitectNow. And I delegate a big chunk of some of my day to day tasks to a team of AI agents.

Here's what I learned. Most people try to build one agent smart enough to do the whole job. That's the trap. The version that actually ships looks more like a team you manage: a handful of narrow specialists, a manager over them, and one agent whose only job is to catch the others lying.

This is the case study, with the real system on screen. How I hired the team, the agent that interrogates the rest before anything ships, the dumb checks I trust more than the smart ones, and where I still keep a human in the loop.

One agent does the work. One agent catches it lying. I'll show you exactly how mine fit together.