--- title: "The Art of Not Checking — Introduction: I No Longer Check What the AI Tells Me" author: garplab publisher: TypingTube license: All rights reserved license_scope: 記事の全体（序論・最終回は CC BY 4.0 の対象外）。引用は法の範囲で自由 canonical: https://typing-tube.net/articles/en/tashikamenai-intro series: "確かめない技術" language: en --- I no longer check what the AI tells me. Not "the tests pass," not "I confirmed it on the screen," not "the fix is complete." And the checking is still done. Before I start. I can stop checking because a mechanism that fails whether or not it is read is already running ahead of me. This series is about building that state one mechanism at a time. A word first on how to read this. The series can be followed by reading alone. Each article starts with what happened on the ground, goes on to what I put down, and ends with what went away. Near the end of each article there is a section called "How to verify." You do not need to read it first. Come back to it when you want to see for yourself why the mechanism you put down is needed. ## Learning the remedy from the symptom Each article starts with a symptom, something that is not working. The remedy comes with the relevant behavior and specifications of the AI, and an explanation of why that remedy fits. **You do not need to learn why it fits before you use it.** Put one mechanism down and the symptom is handled. Even when you know the behavior and the specifications, the AI can still get it wrong depending on how they combine. Ordinary design and development is unlikely to catch these pitfalls, where several behaviors and specifications are tangled together. You learn them from the symptom, one at a time. ## Start by checking what you recognize Start by counting just one thing. The number of times you ran something in your most recent session in order to check. Tests, screenshots, manual runs. Of those, how many had how far you would go decided before you ran them? If you are not building alongside an AI yet, of course there is nothing to count. From here on we go through, one at a time, which part of the checking had not been decided. It reads fine without a count of your own. ## I have written a similar sentence 3 times already "Check your subagents' input, and only that." "Turn down requests to check by eye." "Do not check the work the AI has finished." If you have been reading along, the opening sentence should look familiar. Not checking, and being fine all the same: this is the 4th time that shape has come up. All 3 were about what not to check. This time it is about how to check. ## There is no one left to check This is the fourth series. - Series 1, "[The Art of Not Reading](https://typing-tube.net/articles/en/34b02627c718fa)": it put the AI's output into a shape you do not have to read - Series 2, "[The Art of Not Listening to the AI's Opinions](https://typing-tube.net/articles/en/kikanai-intro)": it put the AI's proposals and judgments into a shape you do not have to weigh up twice - Series 3, "[The Art of Not Telling](https://typing-tube.net/articles/en/iwanai-intro)": it put the human's input into a shape you do not have to say - Series 4 (this one): it puts checking itself into a shape you do not have to decide on the spot Reading the 3 series back, I noticed something. Everything I stopped checking was the thing being checked. The AI's output, the AI's proposals, the production screen, the AI's self-reports. I let go of what to check, one at a time. What was left was how to check. ## Answer yes or no, and the AI fills in the rest Checking looks like a choice between doing it and not doing it. Underneath that hang when, how many times, how far, and by what means. The moment you answer yes or no, someone has to fill in what is underneath. The one who fills it in is the AI that happens to be there. How it gets filled in differs every time. For the same change, the amount that runs varies from day to day. I cannot notice that it varied. That is because what comes back is a report that the checking is done, and what was done, and how much of it, is not written in there. Only the time I spent waiting stretches and shrinks with it. > **Make checking an act, and the AI decides the count, the scope and the means. Decide the checking before the run.** ## The third series ended with a hole left open In the [last article of the main run](https://typing-tube.net/articles/en/iwanai-11-no-review) of the previous series, I wrote this. > What I was short of was not a review. **What I did not know was what the automated tests I had put down were not looking at.** The automated tests passed. A pass only means "nothing is wrong within what was looked at," and yet the report shows the pass alone. What was not looked at appears nowhere. Back then I wrote "you cannot know until you count," and ended the series there. This series starts there. ## This is not a story about going faster This is not a series that lines up tips for cutting the steps of checking. Cutting steps alone takes 1 day, and from that day on the range you are not looking at quietly grows. What it deals with is the other side. Work that makes checking faster hides what you are not looking at. Something that got faster while staying hidden cannot be called checking. So every article puts down, alongside the mechanism that cuts checking, 1 place where the part you cut shows up. A mechanism that does not show it does not go in. The setting is the same as the 3 previous series: [typingtube](https://typingtube.net), a web service for practicing typing along with music videos on YouTube. I run it on my own, and it is live in production. You can start here without having read the 3 previous series. Each article opens with a single line about what the previous series said, and that line is enough. That said, this series is about what comes after you have finished putting down automated tests and hooks. If you have not put them down yet, it is quicker to start from series 1. One more thing to be clear about. This is the record of one person's workplace. On a team of several people, checking is also a report aimed at others, so parts of it are left that you cannot settle by your own convenience alone. ## How the series is laid out - [1. The art of not taking screenshots](https://typing-tube.net/articles/en/tashikamenai-01-no-screenshots) - [2. The art of not trusting the declaration](https://typing-tube.net/articles/en/tashikamenai-02-declaration-guard) - [3. The art of not re-running tests](https://typing-tube.net/articles/en/tashikamenai-03-no-rerun) - [4. The art of not reading false failures](https://typing-tube.net/articles/en/tashikamenai-04-false-failures) - [5. The art of not running everything](https://typing-tube.net/articles/en/tashikamenai-05-selective-e2e) The [finale](https://typing-tube.net/articles/en/tashikamenai-finale) pulls the 5 together and shows how I get by without checking. Each article is finished once you put down a single file or script. Articles 3 through 5 add to the same 1 script in order, so read just those 3 in order. You do not have to put them all down today. Each one opens with a way to check the symptom in your own environment, so pick up the article whose symptom you are seeing. --- Next time, "The art of not taking screenshots." **I stopped having the AI take screenshots.** And broken layouts still get found.