i use the autocomplete. i don’t use my company’s dumb bespoke plugins, but i do use my corporate GitHub token for Copilot. i almost never find agent modes to be less trouble than they’re worth.

finish writing this test? bang out this quick iteration or easy implementation? great.

implement this feature? do this refactor? i feel like i spend more time fixing the subtle mistakes and false assumptions than if i just did it myself. then i can also properly explain and document the work without blindly trusting an albeit impressive heuristics engine.