We have explored copilot and Atlassian’s Rovo agent. Speaking as a novice - the better the prompt you write, the better the chances of getting a decent output. I have found running the same prompt on the same requirements can give you different results, which probably would slow you down!
The approach I’m using now is that I write a set of tests first. Then I use an agent with detailed prompts to generate a set. I’ll go through the list and take from there what I may have missed.
The agent is prompted to look for edge cases, data variations and will search related material to generate regression tests. It halucinates a bit, but I’ve learned to scan the results quickly.
You could argue that you are using AI to save the team time by have a better suite of tests that reduces issues found in UAT or production 