Nina: The 15-minute sprint method

I was talking with my friend Nina, who is a career coach working with people in the SF Bay Area and Taiwan, and because she hears from a great many of them at once, she tends to notice a pattern well before any one of us would. The pattern she has been hearing lately is that the development process around AI has become difficult to sit inside of. It was an accurate description of where I have been. My days had collapsed into one undifferentiated stretch of waking hours, and what wears on me is less the work itself than the shape of it — an instruction goes out, the machine runs for some indeterminate while, and what returns is more thorough than I have the capacity to take in.

Nina: Really — my coaching clients bring this up too. The dev process under AI is oppressive.

Me: You give the AI an instruction and it runs for a while. When it finishes it reports back a pile of things, and you have to read all of it. What it says is too comprehensive.

Nina: Yeah, it’s easy to get overwhelmed.

Me: AI is not like a person who gets tired and starts fobbing you off. It really is easy to be overwhelmed by it.

Nina: The approach my clients worked out is to time-box each stage. Give the instruction, QA, give feedback — then instruction, QA, feedback again. Each one capped at 10 to 15 minutes. When the time is up it’s fine if it isn’t finished; you pick it up on the next round of the same stage. They felt the anxiety of always waiting improved a lot.

Nina: You can also put in another 15 minutes to do some push-ups, let yourself rest and step out of it 😆

Me: Do you mean 15 minutes between each prompt?

Nina: More that you do each kind of work within 15 minutes, and after 15 minutes you jump to the next kind.

Nina: Say you’re coding three projects at once. The first stage is “code & prompt”: in those 15 minutes you code project A for 10 minutes and hit run, then switch to project B for the remaining 5. When the time is up — or once B is running — you move to the next stage, “review result / QA”, for 15 minutes. Then the next stage, “feedback & more prompt”. Something like that.

Me: That really does feel like it would have more rhythm. I’ve been using all my waking hours in full burnout mode, and I’m starting to feel I can’t keep it up. After I improved my own agentic workflow the things the AI gives back are genuinely valuable — every answer, I want to write it down. It is very taxing on the brain.

Nina: Glad it worked out. When you go back to revise, or for future work, you can use this rhythm and turn it into a long game.

Running two or three projects at once was never the missing piece; I had been doing that all along. What I lacked was any edge. Everything advanced at the same time and without pause, in whatever order the reports happened to arrive, which left me permanently in the middle of something and never once at a place where a piece of work was allowed to remain unfinished. Nina’s arrangement puts the boundary somewhere I had not thought to look for it — around the kind of work rather than around the project — and lets the clock, rather than completion, decide when a stage has come to an end.

I am looking forward to trying it on the next thing I build, and to seeing how a day feels with edges in it.