< Twenty Twenty Sprints

My day now runs on a simple cadence: two 20-minute sprints an hour, with a 10-minute gap. It fell out of my AI coding workflow and ended up looking a lot like a pomodoro.

In each block I kick off a set of tasks, nudge them forward, and then "burn down" any threads that can be closed out. If I have any long-running tasks (refactors, upgrades, experiments), I'll set them off to come back to in the next block.

Twenty minutes is long enough to get into something, but short enough to achieve a real outcome. The shape aligns with my intention of optimizing for Single-Prompt Outcomes — I shape each prompt to land as close to "first time" as possible[1].

To make this work I keep a large number of features in flight in parallel. With ~20 open tasks across branches and worktrees[2], there are always a bunch of plates spinning. By the time I finish a sprint on one thread, another has produced something worth looking at.

Sprint, rest, sprint, rest

In the break I step away from the computer. This is time for a walk, cleaning the kitchen, coffee, exercise, calling a friend[3]. Even if I "work", I'll do it in my notebook on paper.

I do miss the "deep work", but this approach produces a more balanced day. I'm not "beast mode" programming. I'm not smashing things out, headphones on, for hours in a stretch. Instead, I'm kicking off a thread and then casually thinking it over while I clean the kitchen. Usually I come back to completed work, or with a clearer head on what needs to happen next.

Here are some keys I've found to make this work:

Also read: My Current AI Coding Workflow and Sharpening the Axe (Branch).

I write about AI, organizations, and engineering leverage: find out about me and subscribe here.

Discuss and share via the meta page . Filed under AI, Productivity, and 100PR.

Footnotes

  1. I rarely use PRDs. I'm doing design, product, and engineering in one, so that's a luxury in some senses. I use Battleship Prompts to give a good chance I'll get an outcome from most prompts. If something doesn't land, I throw it away.

  2. From my current AI workflow -- ~20 open tasks is my sweet spot for balancing parallelism and WIP.

  3. Yes, sometimes this exceeds 10 minutes, that's fine. I usually am able to get back on pace with minor adjustments along the way.