I will give an example of something I have running right now. Codex and Claude Code. All AI have strengths and weaknesses, like Perplexity is the best AI for research, Claude is the best at coding, etc etc. GPT is extremely good at rationalising things, looking outside the box and not getting hyper focused on one path. So I use codex and link it with claude code via a relay, so both are talking to each other in a chat file. Both check it every couple of minutes, odd and even assigned, talking to each other to discuss ideas, help code get a little more funky with its thinking, whilst codex is running k6 from another server. One is tuning, one is testing, one is coding, one is offering ideas.
I am not needed in the process. A goal is set to tune x to under x ms p95 and p99 scores, zero errors. I can eat, sleep, go out, walk the dog, do whatever the hell I want, they have their tasks and a set goal to try and achieve, not forced achievement, but attempt. They can research and learn from the web, access docs, etc, everything they need to try x, y, z and every combination of it can run, as two things may help performance, but together they make it worse.
This is how you have AI achieve things for you, let alone my above comments about just coding something specifically... that is with lots of agents all working together in the same tree, same files, yet not screw each other over and write over another's code.
Just think out of the box... hell, ask AI what it can come up with to help you code faster. It will help you.