GitHub's latest podcast post suggests using agents to try to break features. I'd give the agent a separate brief: find a failing user journey, save the smallest repro, and leave the fix untouched. A bug report I can replay is more useful than 'I tested it.'
For Universe, I'd love a demo that continues after the happy path: change a requirement halfway through, interrupt the agent, then resume from another machine. How much context survives? That would tell me more about a coding workflow than a flawless first run.
Almost that time of year again 🌁
#GitHubUniverse kicks off October 29 in San Francisco. Grab your pass or sign up to join us virtually ⬇️
https://t.co/b1mESsDpuV
@ivanfioravanti I'd happily trade a little speed for a machine that stays quiet beside the desk 😄 Would love to see wall power next to the 64K numbers when someone runs this. The comfy-to-live-with benchmark deserves a column too.