For me, the act of coding felt the most blissful when I reached the flow state.  With agenting coding my speed, of course, has increased. 
But I never get to flow.

It brings up the question as to why. If I take a step back, even for coding I had to switch context.
Write some code, build, check the terminal output, do stuff, run tests, etc. The overall act of building a feature till the git push did have quite a few different steps. It was not as if I was always typing in my flow state and staring at the screen with just a single flow of thought.
But somehow doing the same with agent or agent-specific coding feels much more jarring.

I am never too happy about doing agentic coding. I'm never satisfied after doing it. The end product looks super awesome, and it is typically much faster. There is a wow factor to it. That's about it. It feels like having KFC vs a satiating full course meal. 

And I think I have finally been able to figure out why.  
When I was doing things manually before, i had a sense of time and predictability. Yes, I had to run a build, or I had to run a test, but I knew approximately how long it would take. It might take seconds; it might take even a minute. But with an agent, I am never sure. Sometimes a prompt can take one minute. Sometimes a prompt can take ten minutes. Sometimes it doesn't work, which is basically staring into the void of what the prompt is doing and  steering it from time to time without any idea of preedictability. 

I don't know how I can achieve flow with this, with these types of unpredictable intervals. Maybe I will, maybe I won't, but for now it does feel deeply unpleasurable. It is akin to coding while hanging upside down. 

Date: 2026-02-27