{"course":{"slug":"prompting","title":"Getting Real Work out of a Model","url":"https://addaly.com/learn/prompting","level":"beginner","track":"start"},"lesson":{"slug":"knowing-when-to-stop","title":"When to stop prompting and do it yourself","url":"https://addaly.com/learn/prompting/knowing-when-to-stop","position":54,"of":72,"minutes":8,"answer":"Decide beforehand how long the task would take by hand, stop when the same error survives two genuine fixes, and hand back the parts that need judgement, accountability or knowledge only you have.","keyIdea":"Decide beforehand how long the task would take by hand, stop when the same error survives two genuine fixes, and hand back the parts that need judgement, accountability or knowledge only you have."},"body":{"format":"markdown","text":"## The loop that eats an afternoon\n\nTwenty minutes in, the answer is close. Thirty minutes in, it is differently close. At fifty minutes you are rewriting a prompt for a task you could have finished by hand in fifteen.\n\nEverybody does this. It is not stupidity; it is the shape of the activity. Each attempt is nearly right, so one more seems reasonable, and there is no natural stopping point. The same structure keeps people at slot machines.\n\n## Set the budget first\n\nBefore starting, decide how long you would spend doing it yourself. That number is your budget. When you reach it, stop and do it.\n\nThis one habit saves more time than every technique in this course, and it costs nothing but the discipline of saying the number out loud beforehand.\n\n## The four signals to stop\n\n**1. The same error survives two genuine fixes.** Not two rewordings: two real changes, addressing different causes. If the citation is still invented after you pasted the source and required a quote, you have hit a limit of the approach.\n\n**2. You are correcting the same thing repeatedly.** You said \"no bullet points\" three times and bullets keep coming back. Either the correction is competing with something else in the context, or you have found a strong default. Start fresh once with the instruction at the top; if it recurs, do that part yourself.\n\n**3. Specifying the answer has become the work.** By the fourth iteration you have described what you want in such detail that you have effectively written it. This is worth noticing rather than resenting: it usually means the task was mostly judgement, and judgement is not what you were delegating.\n\n**4. You cannot tell whether it is right.** The most important one. If checking the output requires the expertise the output was supposed to supply, you are not saving effort; you are moving it and adding a verification problem.\n\n## Partial handover is the usual answer\n\nStopping rarely means abandoning the tool. It means splitting the job at the right place.\n\n- It drafts, you write the two paragraphs that carry the argument.\n- It extracts forty rows, you check the eight flagged `uncertain`.\n- It lists the options, you make the call.\n- You write the difficult message, it checks the tone.\n\nThe parts that need judgement, accountability or knowledge only you have are the parts to keep. That is not a defeat. It is what the division of labour looks like when one party has no stake in the outcome.\n\n## When the tool is wrong for the job\n\nSome tasks are structurally unsuited, and no prompt fixes a structural mismatch:\n\n- **Precise arithmetic over many rows.** Use a spreadsheet or code.\n- **Anything needing current facts** without a search tool attached.\n- **Anything about your private systems** (your database, your files, your bookings) without a retrieval path to them.\n- **Anything where you need a guarantee** rather than a good-enough answer. There is no prompt that produces a guarantee.\n- **Anything where being wrong is expensive and unverifiable.** Combining an expensive error with an unverifiable answer is the one combination to refuse outright.\n\n## The self-honest question\n\nAt the end of a long session, ask what you actually got. Sometimes the honest answer is: a worse version of what I would have written, arrived at more slowly, plus a feeling of productivity.\n\nThat answer is not a reason to stop using these tools. It is the information that tells you which tasks they are for, and you only get it by asking, because the sense of progress inside the loop is not a measurement.\n\n## The rule\n\n**Set the budget before you start. Stop when the same error survives two real fixes. Keep the part that needs you.**\n\nThree sentences, and they are the difference between a tool that saves you an hour a day and a habit that costs you one.\n\n## A note on the skill you are keeping\n\nThere is a version of stopping that is not about time at all. Some tasks are how you stay good at your job: writing the difficult analysis, drafting the argument, working the problem through.\n\nHanding those over saves an afternoon and costs something slower to notice. Nobody can tell you which tasks those are for you, and it is worth deciding deliberately rather than discovering in two years that a skill quietly went. The tools are extremely good at the work you find tedious. Some of what you find tedious is practice."},"check":{"question":"Which situation most clearly means stopping and doing the task yourself?","options":[{"text":"Checking whether the answer is right would require the same expertise the answer was supposed to provide.","correct":true},{"text":"The output needs three more rounds of correction to be usable.","correct":false,"tempting":"Might be fine if each round is producing real progress and the budget has not been reached."},{"text":"The model keeps producing bullet points despite instructions to use prose.","correct":false,"tempting":"A strong default worth one fresh start with the instruction placed first; not yet a reason to abandon the task."},{"text":"The task is taking longer than expected and involves specialised terminology.","correct":false,"tempting":"Specialised vocabulary is a context problem, usually fixed by pasting a glossary rather than by abandoning the attempt."}]},"prev":{"slug":"asking-it-to-fix-the-prompt","title":"Getting the model to improve your prompt","url":"https://addaly.com/learn/prompting/asking-it-to-fix-the-prompt"},"next":{"slug":"spot-a-guess","title":"How to tell when it is guessing","url":"https://addaly.com/learn/prompting/spot-a-guess"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'When to stop prompting and do it yourself', in Getting Real Work out of a Model. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/prompting/knowing-when-to-stop (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"When to stop prompting and do it yourself","courseTitle":"Getting Real Work out of a Model","module":{"index":6,"slug":"from-lucky-to-reliable","title":"From lucky to reliable"},"answer":null,"keyIdea":"Decide beforehand how long the task would take by hand, stop when the same error survives two genuine fixes, and hand back the parts that need judgement, accountability or knowledge only you have.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-loop-that-eats-an-afternoon","runs":[{"text":"The loop that eats an afternoon"}]},{"t":"p","runs":[{"text":"Twenty minutes in, the answer is close. Thirty minutes in, it is differently close. At fifty minutes you are rewriting a prompt for a task you could have finished by hand in fifteen."}]},{"t":"p","runs":[{"text":"Everybody does this. It is not stupidity; it is the shape of the activity. Each attempt is nearly right, so one more seems reasonable, and there is no natural stopping point. The same structure keeps people at slot machines."}]},{"t":"h2","id":"set-the-budget-first","runs":[{"text":"Set the budget first"}]},{"t":"p","runs":[{"text":"Before starting, decide how long you would spend doing it yourself. That number is your budget. When you reach it, stop and do it."}]},{"t":"p","runs":[{"text":"This one habit saves more time than every technique in this course, and it costs nothing but the discipline of saying the number out loud beforehand."}]},{"t":"h2","id":"the-four-signals-to-stop","runs":[{"text":"The four signals to stop"}]},{"t":"p","runs":[{"text":"1. The same error survives two genuine fixes.","b":true},{"text":" Not two rewordings: two real changes, addressing different causes. If the citation is still invented after you pasted the source and required a quote, you have hit a limit of the approach."}]},{"t":"p","runs":[{"text":"2. You are correcting the same thing repeatedly.","b":true},{"text":" You said “no bullet points” three times and bullets keep coming back. Either the correction is competing with something else in the context, or you have found a strong default. Start fresh once with the instruction at the top; if it recurs, do that part yourself."}]},{"t":"p","runs":[{"text":"3. Specifying the answer has become the work.","b":true},{"text":" By the fourth iteration you have described what you want in such detail that you have effectively written it. This is worth noticing rather than resenting: it usually means the task was mostly judgement, and judgement is not what you were delegating."}]},{"t":"p","runs":[{"text":"4. You cannot tell whether it is right.","b":true},{"text":" The most important one. If checking the output requires the expertise the output was supposed to supply, you are not saving effort; you are moving it and adding a verification problem."}]},{"t":"h2","id":"partial-handover-is-the-usual-answer","runs":[{"text":"Partial handover is the usual answer"}]},{"t":"p","runs":[{"text":"Stopping rarely means abandoning the tool. It means splitting the job at the right place."}]},{"t":"ul","items":[[{"text":"It drafts, you write the two paragraphs that carry the argument."}],[{"text":"It extracts forty rows, you check the eight flagged "},{"text":"uncertain","code":true},{"text":"."}],[{"text":"It lists the options, you make the call."}],[{"text":"You write the difficult message, it checks the tone."}]]},{"t":"p","runs":[{"text":"The parts that need judgement, accountability or knowledge only you have are the parts to keep. That is not a defeat. It is what the division of labour looks like when one party has no stake in the outcome."}]},{"t":"h2","id":"when-the-tool-is-wrong-for-the-job","runs":[{"text":"When the tool is wrong for the job"}]},{"t":"p","runs":[{"text":"Some tasks are structurally unsuited, and no prompt fixes a structural mismatch:"}]},{"t":"ul","items":[[{"text":"Precise arithmetic over many rows.","b":true},{"text":" Use a spreadsheet or code."}],[{"text":"Anything needing current facts","b":true},{"text":" without a search tool attached."}],[{"text":"Anything about your private systems","b":true},{"text":" (your database, your files, your bookings) without a retrieval path to them."}],[{"text":"Anything where you need a guarantee","b":true},{"text":" rather than a good-enough answer. There is no prompt that produces a guarantee."}],[{"text":"Anything where being wrong is expensive and unverifiable.","b":true},{"text":" Combining an expensive error with an unverifiable answer is the one combination to refuse outright."}]]},{"t":"h2","id":"the-self-honest-question","runs":[{"text":"The self-honest question"}]},{"t":"p","runs":[{"text":"At the end of a long session, ask what you actually got. Sometimes the honest answer is: a worse version of what I would have written, arrived at more slowly, plus a feeling of productivity."}]},{"t":"p","runs":[{"text":"That answer is not a reason to stop using these tools. It is the information that tells you which tasks they are for, and you only get it by asking, because the sense of progress inside the loop is not a measurement."}]},{"t":"h2","id":"the-rule","runs":[{"text":"The rule"}]},{"t":"p","runs":[{"text":"Set the budget before you start. Stop when the same error survives two real fixes. Keep the part that needs you.","b":true}]},{"t":"p","runs":[{"text":"Three sentences, and they are the difference between a tool that saves you an hour a day and a habit that costs you one."}]},{"t":"h2","id":"a-note-on-the-skill-you-are-keeping","runs":[{"text":"A note on the skill you are keeping"}]},{"t":"p","runs":[{"text":"There is a version of stopping that is not about time at all. Some tasks are how you stay good at your job: writing the difficult analysis, drafting the argument, working the problem through."}]},{"t":"p","runs":[{"text":"Handing those over saves an afternoon and costs something slower to notice. Nobody can tell you which tasks those are for you, and it is worth deciding deliberately rather than discovering in two years that a skill quietly went. The tools are extremely good at the work you find tedious. Some of what you find tedious is practice."}]}],"check":{"question":[{"text":"Which situation most clearly means stopping and doing the task yourself?"}],"options":[{"text":[{"text":"Checking whether the answer is right would require the same expertise the answer was supposed to provide."}],"correct":true,"tempting":null},{"text":[{"text":"The output needs three more rounds of correction to be usable."}],"correct":false,"tempting":[{"text":"Might be fine if each round is producing real progress and the budget has not been reached."}]},{"text":[{"text":"The model keeps producing bullet points despite instructions to use prose."}],"correct":false,"tempting":[{"text":"A strong default worth one fresh start with the instruction placed first; not yet a reason to abandon the task."}]},{"text":[{"text":"The task is taking longer than expected and involves specialised terminology."}],"correct":false,"tempting":[{"text":"Specialised vocabulary is a context problem, usually fixed by pasting a glossary rather than by abandoning the attempt."}]}],"order":[1,0,3,2],"why":null},"prev":{"slug":"asking-it-to-fix-the-prompt","title":"Getting the model to improve your prompt","minutes":8},"next":{"slug":"spot-a-guess","title":"How to tell when it is guessing","minutes":9}}}