We tried Gemini 2.0 and it's good at bug reporting
It even writes repro steps (and adds timestamps to link each repro step to the exact moment in the session recording playback). Plus, it's a lot more concise than most people reporting bugs, so engineers can understand immediately what is going on without having to read a lot of text or decipher a confusing ticket :)
You can see the demo here: https://x.com/jamdotdev/status/1887523689518014891
One cool feature: once you `tab` to accept Gemini's AI-written bug report, it will log it right in your issue tracker (so you never have to open Jira to log a Jira again). (We integrated with Jira, Linear, and 7 others)
Why Gemini 2.0? We initially chose Gemini because of its native ability to process videos (which other LLMs don’t have, so it would require a lot less pre-processing for us and avoid adding a small amount of latency). And then we enabled Gemini 2.0 while it was in experimental mode. It turns out Gemini 2.0 is pretty good at logging bugs. It's fully writing 27% of the early access users bug reports, which we are pretty happy to see.
You can try it out for yourself: jam.dev/ai I’d love to hear your thoughts, my email is dani@jam.dev – reach out anytime. We want to make this a lot more fully featured – and would love your input where you'd like to see this go.
Definitely we think humans have a major role to play in testing software, but compiling the debug logs and writing up the summary and repro steps for engineers seems like something AI excels at and people are a lot more imprecise at (and that imprecision causes a lot of wasted effort today on engineering teams).
By the way, we’re hiring engineers and if you're really into AI-powered devtools and debugging, would love to chat: jam.dev/careers