Zen and the Art of Getting Out of Your Own Way
Say Nothing First
At work, some of my best sessions with Claude begin with me saying almost nothing. I hand over a document and tell it to look at it. No task, no question, no format.
Claude tells me what it is, summarizes it, or pulls out the key points, and then asks what I want to do. That question is the whole trick. By the time I answer, I have seen the document through someone else's eyes, and I usually know better what I need than I did a minute earlier.
The Half-Effort Prompt
A lot of my prompts are, to put it politely, half-effort, and they still produce very good results. I used to treat that as a quirk. Now I think it is closer to a method.
Three things seem to be doing the work:
- Context. A loose nudge lands well when Claude already knows what I am working on. A short prompt inside a rich context is a different thing from a short prompt cold.
- Cheap iteration. The first answer is a draft, not a verdict. One correction costs me a sentence, so there is little reason to agonize over the opening move.
- Taste. I know a good result when I see one. That judgment is the real skill, and the casual prompt only looks casual because the judgment sits somewhere else.
Why Boxing It In Backfires
In the past, I noticed that boxing in a language model could lead to hallucinations, or at the very least to simple mistakes. Hallucinations happen less now, but I also avoid many of them by leaving requests somewhat open.
My working theory is that a tight demand can ask for more than the material can supply. Ask for ten supporting quotes from a set of documents that holds six good ones, and the pressure to deliver ten becomes the problem. An open ask lets the answer be as big or as small as the material allows.
So I box in the form and leave the content alone. "Keep it short" shapes the output. "Find the three flaws" assumes there are three.
What I get most often from an open request is a different interpretation than the one I would have made. That is fair, and I have come to value it. A different reading shows me what I assumed without noticing, and now and then it is simply better than mine. Someone who prompts tightly never sees those, because the box filters them out.
Where the Structure Lives
Getting out of your own way does not mean switching off judgment. It means moving judgment to where it works best.
Most advice puts the guardrails up front: precise instructions, strict constraints, a long prompt that tries to prevent every failure in advance. I put them afterward. The generating is open and loose. The checking is careful, and it is mine.
If I box in the draft stage, I get the model's best guess at my constraints. If I leave it open and review afterward, I get a range of possibilities and then choose among them. That is how the surprises get through.
The Writing Routine
My writing process with Claude is more structured than how I use it at my job, but it has no guardrails in the usual sense. It developed over time, the way a path gets worn by walking. Nobody planned the route, but it is there because it works.
- Brainstorm together.
- Research together.
- Work on an outline, sometimes together, sometimes on my own.
- Create a rough draft, sometimes dictating to Claude, sometimes doing a back and forth, sometimes letting it drive just to get words on paper.
- Then I sit with it for a few hours to a day, tweaking boards and ideas and moving things around.
- Check the claims that need checking.
- Make sure it's really my point of view, and that I agree with all of it.
- Run it by a friend, occasionally.
The loose part is the collaboration. The careful part is everything after the draft, and none of it is automated. Two steps carry most of the weight. The day spent with a draft is where an answer turns into something I actually understand. And making sure it's authentically me catches the things that don't feel right, which fact-checking would never flag.
The Question You Didn't Quite Ask
I wrote recently that answers are not understanding. A fluent response is still only a response until someone has turned it over and decided what they think of it. That is why I can afford to be loose at the start: the understanding was never going to come from the prompt.
Getting out of your own way, as I practice it, is a kind of trade. I give up control at the start and take it back at the end. Claude answers the question I did not quite ask, and the gap between that answer and the one I expected is usually where I learn something. The less I grip at the beginning, the more the ending depends on me: the day with the draft, the checking, the question of whether it's all me.
So the next time you have a document and no idea what to do with it, try handing it over and saying nothing. See what it asks. Then take your time with the answer.