focus more, multi-task less
Most of my way of working has changed with AI (LLMs) agents.
I don't write code myself anymore; the agents do. I craft prompts, guide them, and structure the designs. It's quicker than me writing the code, so as long as the steer is good, I've got higher output in that sense.
With the rise of that workflow change, everybody (random people on Twitter) was saying that you need to orchestrate lots of agents at once - 'you need to be running at least 5 at any point now, otherwise you're not getting enough value from them. Just keep switching between them while they're busy'
I was doing that. I was looking at things like Gas town to help manage it. I was hitting the 10+ agents-at-once threshold.
But I realised, I wasn't building stuff with quality and speed. I probably wasn't really moving much faster. Context switching and management were taking so much time. I felt like I was, but I wasn't.
Now? I'm trying to cap myself at 3 at a time. Or, I'll pause one. I try to think more about my headspace and context switching. Finish an agent's tasks/goals before opening a new one, unless it's to discover/answer some questions, which will be short-lived.
I'm building features quicker than before, and with the same, if not better, bar for quality and automated testing. I'm big on TDD, but it can be a bit slow-moving at times; this allows me to do it at speed.
Do less to do more. Slow down to speed up.