Focus
Pointing a now-unlimited output at the one thing that matters.
The behaviour, in one line: I aim my AI-amplified effort at one clear outcome and stop when it’s good enough, instead of generating endless output that never moves the goal.
A firehose and a nozzle move the same water, and one of them puts it where you wanted it. Now that output is nearly free, volume was never going to be the constraint. Aiming it is.
Producing things used to be the expensive part, so being able to produce more was straightforwardly good. That constraint has gone. You can now generate code, copy, designs and analysis faster than you can read them, which quietly replaces the old question — can I make this? — with a harder one: is this worth making? Focus is what answers the second question. Without it you produce more than ever and move less.
| What it is | What it isn’t |
|---|---|
| One outcome, written first | Whatever the last prompt returned |
| A stopping bar set in advance | Stopping when you get tired |
| One thread you can steer | Ten you cannot evaluate |
| Counting what shipped | Counting what was produced |
What it looks like. You could generate twelve more variants in the time it takes to read this. You do not. What is on screen clears the bar you wrote down an hour ago, so you send it and start the next thing.
- Any tool that makes output free. Cheap production removes the natural stopping point. Something has to replace it, and it will not appear on its own.
- The busy two weeks with nothing in it. Prompt, wait, prompt, wait feels like work and produces real artefacts. It can still move nothing.
- Delegating to agents. Running several at once only helps if you can tell which output was good. Otherwise you have multiplied the reading, not the work.
- Write the one outcome before you generate anything, and judge every output against it rather than against how much you made.
- Set the “good enough” bar in the same breath, so stopping exists in advance instead of arriving by exhaustion.
- Steer one thread at a time. Ten parallel tasks you cannot evaluate is not ten times the work.
- Count what shipped, not what was produced. The two diverge quickly, and only one of them is the job.
- Break it into parts
Four parts: writing the one outcome, setting a “good enough” bar, steering a single thread, and counting value rather than volume.
- The part that carries it
Writing the single outcome before you start generating. Without it, cheap unlimited output has nothing to be judged against.
- The order to learn it in
- Write the one outcome first — before any prompt, any draft, any build.
- Set the stopping bar in the same breath, so “done” exists in advance rather than by exhaustion.
- Then hold to one thread at a time, and start counting what shipped instead of what was produced.
- Put something at stake
Track for two weeks what you actually shipped against what you generated, and show someone the ratio.
- Compress it
One outcome, written first. One stopping bar. One thread. Count what shipped.
- The smallest useful dose
One written outcome and stopping bar per work session that involves AI.
- Anchor it to something you know
A firehose and a nozzle. The volume of water was never the problem.
“It is not enough to be busy; so are the ants. The question is: what are we busy about?”
— Henry David Thoreau
The interactive version is in the course. The diagram, the simulation and the four-question check for Focus run in the free course, where your progress across all 24 skills is kept.
Not sure where you stand? Take the twelve-minute self-assessment, or see the 30-day course that turns these skills into a roadmap.