Multi-threading my work
When I’m at my computer I tend to have a primary task and a secondary task. That’s because my primary task is mediated by Claude or OpenAI, and my secondary task is what I do while the AI is munching on tokens.
Right now, my secondary task is writing.
I’ve been setting up OpenClaw to become a useful agent-provider for my family and for work. As my wife will no doubt tell you, this has been an enjoyable black hole, because it hits all of my buttons: I love geeky things, I love setting things up, I love solving puzzles. I’ve been up late in the evening getting APIs configured and doing things like connecting OpenClaw to a local AI model running on a different computer on my home network (which is stupider, but infinitely cheaper than the Anthropic API).
The entire process has been assisted by AIs. Basically I point at things and say “configure this” or “build a way to connect to the Monday.com API” and then iterate. I’m simplifying — but the point is that while the agent does that work, I’m flipping to something else. The temptation to proliferate is real. I think one evening last week I was cycling between four tasks: setting up OpenClaw, vibe coding a video game (I’ll share that one later), updating my neovim package manager, and writing fiction.
After a few hours of this, I’d be dog tired but completely stimulated. It was impossible to go to sleep; and I’m not convinced I made better progress on any of these threads than I would have if I tried to do them in sequence.
AI creates the temptation to multi-thread because it feels wasteful to sit there and wait for the AI to process a batch of work. Over time, this will become less of an issue because the more capable AI tools will be able to run for longer times independently. Right now I tend to get up to 10-15 minutes of independent work on well structured things before I have to spin the plate again. But in the future I can imagine setting out 2-3 hours worth of work.
For now, multi-threading is draining and it reduces my throughput. So I’ve settled on restricting myself to two active streams of work at once: primary and secondary. The reason I distinguish between primary and secondary is so that I know which task interrupts the other: I will divert from writing when it’s time for me to check in on my primary task.