What's this about?
dotey reports that AI is now very good at performance work: screenshot Activity Monitor, and from the PIDs alone the agent quickly identifies the cause of high memory use. He contrasts it with the old process — capture several memory dumps, analyse the stacks with tooling, and work hard for a probable cause. His job is now picking a few scenarios to test.
What exactly changed?
Not the analysis but the entry cost. Memory profiling required knowing which tool to run, how to trigger a dump at the right moment, and how to read the output — a specialist skill that made most developers avoid it. Starting from a screenshot removes every one of those steps, so the investigation happens at all. As he notes, he doesn't actually know how it does it, which is its own observation.
What's the community saying?
- Recognition from people who remember how tedious this workflow was, which is most of the audience.
- His forward-looking note that in principle the agent could drive the whole loop itself with computer use, including running the test scenarios.
- The honest caveat embedded in his own post — not understanding the method makes verification harder, which matters more for performance claims than for most tasks.