"We need a chatbot for our customer tickets."
The chatbot wasn't the first problem. Tickets arrived unsorted, so anything answering them would have been guessing. We built a layer to sort and understand tickets, then one that asks customers the right follow-up. The chatbot came later, and it worked.
"Here's the architecture we researched ourselves."
They were subject-matter experts, not AI experts. We worked through their plan with them rather than around them. Two years later, they're still with us and still getting results.
"We built 36 agents and they won't scale."
The agents weren't the problem. Their data and knowledge layer was never built to carry that load. They weren't happy to hear it, and we didn't end up working together.