- Published on
What I do while AI writes my code
- Authors

- Name
- Leon
- @leonpzhao
I used to get absorbed in writing code. There was a steady feedback loop: I'd make progress, think through the next problem, and keep going. Finishing the work felt satisfying.
Now I hand most coding tasks over to AI. In doing so, I've also handed over the long stretches of work I could lose myself in. What I get back is time broken into pieces.
I give AI a task, and it might take twenty minutes to finish. That leaves me with twenty minutes to fill.
I used to handle that wait in two ways.
The first was to run several tasks in parallel. I'd start another task while the first was running. Just as I finished explaining the new one, another would have results ready. I'd review the implementation, point out problems, and ask for changes, then move on to the next task. Once enough tasks were in flight, all that switching wore me out.
The second was to do something else, like browse the web or scroll through Twitter. But taking in information in little bursts left me feeling surrounded by it, always chasing whatever was newest. I'd come away tired, with little to show for what I'd read.
Lately, I've been trying to put some boundaries around that waiting time and spend it on something I can stay with, something that requires me to do the thinking.
One of those things is writing. I built an app to collect stray thoughts. When I have a spare stretch of time, I connect those notes and work them into an article. This piece came together across several of those stretches.
The other is reading a long essay. I split my screen, with Codex's progress on one side and a Paul Graham essay on the other. Lately, I've been reading How to Do Great Work. A long essay gives my attention somewhere to settle. I can follow it for a while, and the reading leaves me with new thoughts of my own. That feels satisfying.
What matters to me is doing one thing at a time, without constantly asking my attention to switch contexts.
I often see people emphasize how many AI tasks they manage at once and how much work they get done. I'm starting to question how efficient that really is. When I juggle that many tasks, the cost is my attention, and my thinking becomes shallower.
Does working that way actually help in the long run?