What a failed demo taught our group
Our dispatch agent looped in front of everyone. It was the best week of the Lab.
We tested on clean data, then the live demo met a real delivery sheet and the agent called the same tool eleven times.
What we changed
We test on the messiest file we can find.
Every agent gets a step limit and a way out.
One person tries to break it before every demo.
Nobody remembered the loop a week later. They remembered that we fixed it in two days.
- Agents
- Teamwork
Related writing
BlogI built my first voice agent in a weekend
What broke, what surprised me, and why latency matters more than the model.
BlogA simple way to test your AI tool
Ten real examples and one honest score tell you more than any demo.