But ofc, that annuls much of the time saved using an agentic workflow. #programming #agents #AI
Also, this is one workflow at a time. I don't get the talk of parallelization of agents. How on earth do you deal with all the context switching? I think it would kill me tbh.
It takes all of my cognitive bandwidth just reading and understanding the outputs.
@marytzu
LOL, whenever I tried recently, after a short while I thought, fuck it, rather than coerce a more than stubborn "junior developer", aka ai, into implementing what I want, I just write it myself.
Boring lookup jobs like is it string.join(list) or list.join(string) are good for ai, but writing code is still much quicker than writing an elaborate spec in natural language precise enough to get the intended result.
@HaraldKi yes, I had an example of that recently. I needed to change a hardcoded enum in the legacy code, I knew what I wanted, and it was just easier to make the change myself rather than having to explain it in natural language.
Re: "junior developer": in some ways it's so much smarter (eg creating an ffbd of the code way faster than i could), in either ways it is soo stupid. Like my claude opus 4.6 can't stop using magic strings/numbers no matter the prompt.