What it means
QA (quality assurance) is the structured sweep before launch: run the eval set one more time, exercise every workflow path, check the integrations, verify the logs are landing, confirm the rollback plan works. The technical team does QA.
UAT (user acceptance testing) is what happens after QA passes. The actual end users (the customer service team, the sales reps, the booking staff) use the AI agent on real workflows for one or two weeks. They flag what feels off, even when nothing is technically broken. UAT catches the things QA cannot.
Why it matters
An AI deployment that skips UAT often fails at adoption. The team gets the new tool, finds three things they hate about it, and quietly stops using it. The thing they hated was usually fixable in an afternoon if anyone had asked.
UAT is also how you build internal champions. The first three people who used the AI agent during UAT and helped shape it become the people who teach everyone else. That makes the rollout faster and friendlier.
Example
A property agency runs two weeks of UAT with their three senior agents. They flag: the agent's replies are too long, it does not handle Mandarin enquiries well, and the calendar links break on iOS. All three get fixed before general rollout. By month two, the same three agents are running training sessions for the rest of the team.