{"course":{"slug":"choosing-your-tools","title":"Choosing and Using the Tools","url":"https://addaly.com/learn/choosing-your-tools","level":"beginner","track":"work"},"lesson":{"slug":"hidden-costs","title":"The costs that are not on the invoice","url":"https://addaly.com/learn/choosing-your-tools/hidden-costs","position":36,"of":76,"minutes":8,"answer":"The invoice is the small number: verification time, rework, learning and automation bias decide whether a tool pays.","keyIdea":"The invoice is the small number: verification time, rework, learning and automation bias decide whether a tool pays, and the returns are largest exactly where you can check an answer faster than you could produce it."},"body":{"format":"markdown","text":"## The invoice is the small number\n\nA tool that costs $20 a month can easily cost you far more than $20 a month. The costs that do not appear on any invoice are real, they are often larger, and they are the reason two people with the same subscription have completely different experiences of whether it was worth it.\n\n## Verification time\n\nThis is the big one. If a tool produces a draft in thirty seconds that would have taken you twenty minutes, but you then spend twenty-five minutes checking it because you cannot afford to be wrong, the tool has cost you five minutes.\n\nWhether verification is expensive depends on a property of the task, not of the model: **how long does it take to check an answer compared with producing it?**\n\n- **Cheap to check.** Code that either runs or does not. Arithmetic you can recompute. A translation you can read. A summary of a document in front of you. Here the tool is nearly pure gain.\n- **Expensive to check.** A legal citation, a factual claim about a subject you do not know, a figure taken from a long report, a recommendation whose consequences appear months later. Here the verification can cost more than the work.\n\nThe practical consequence: these tools give the largest genuine returns on tasks where you can verify quickly, and the largest illusory returns where you cannot. Notice which kind you are doing.\n\n## Rework and the confident wrong answer\n\nA wrong answer you catch costs you the time to catch it. A wrong answer you do not catch costs whatever it costs: a supplier paid twice, a clause misread, a patient given the wrong advice, a customer told something untrue.\n\nYou cannot price this directly, but you can price it roughly by asking what the last mistake in this area cost, and how often it would have to happen to wipe out the time saved. For work where being wrong is expensive, that arithmetic frequently says use the tool for drafting and never for deciding.\n\n## Learning and switching\n\nGetting good with a tool takes ten to twenty hours spread over a few weeks. That is a real cost, paid once, and it is why switching tools every month is expensive even when each individual tool is free. You are permanently in the beginner phase.\n\nIt is also why the six-month review cadence is right. Frequent enough to notice a real change, infrequent enough that you keep the skill you built.\n\n## Cognitive costs, stated honestly\n\nTwo of these are documented and worth knowing about.\n\n**Skill maintenance.** If you stop doing a thing yourself, you get worse at it. This matters where the skill is the point (a student learning to write, a junior developer learning to debug) and matters much less where it is not, such as formatting a spreadsheet. The evidence on this is still developing and the effect size is contested. What is not contested is the mechanism: practice maintains ability, and outsourcing practice removes it.\n\n**Automation bias.** People accept a machine's answer more readily than a colleague's, and check it less. This is a well-established finding across decades of research into other automated systems, and there is no reason to think it does not apply here. It is worth designing around: decide in advance which categories of answer you always verify, so the decision is not made in the moment when the answer looks convincing.\n\n## The costs that go the other way\n\nBeing fair, some hidden costs are negative: the tool saves you things that never appear on an invoice either. The email you were dreading that now takes four minutes. The document in a language you read slowly. The code you did not have to ask a busy colleague about. The blank page you did not have to face. For a great many people these are the actual value, and they are just as invisible to the accountant as the verification time.\n\n## Putting a number on it\n\nFor two weeks, keep one line per day: *minutes saved, minutes spent checking, one mistake nearly made*. It is ninety seconds a day and it produces a real answer.\n\nMost people find the tool is worth substantially more than they pay on the tasks they can check, and worth much less than they thought on the tasks they cannot. That is a useful thing to know about your own work, and no review will tell you."},"check":{"question":"Two colleagues use the same assistant. One saves several hours a week; the other finds it costs time. Their tasks differ in one property. Which?","options":[{"text":"How complex their tasks are, since the model handles simple work well and struggles with complex work.","correct":false,"tempting":"Sorts tasks by difficulty when the deciding property is checkability; hard code that either runs or does not is cheap to verify."},{"text":"How long their prompts are, since the one saving time has learned to give more context.","correct":false,"tempting":"Credits prompt skill, which changes answer quality without changing how long verification takes."},{"text":"How long checking an answer takes against producing it, which decides if verification eats the saving.","correct":true},{"text":"How much of the work is drafting rather than deciding, since drafting is what these tools are designed for.","correct":false,"tempting":"Restates a sound guideline rather than the mechanism, and drafting a claim you cannot verify is still expensive."}]},"prev":{"slug":"three-billing-shapes","title":"Subscriptions, credits and pay-as-you-go","url":"https://addaly.com/learn/choosing-your-tools/three-billing-shapes"},"next":{"slug":"paying-from-anywhere","title":"Paying for it from where you live","url":"https://addaly.com/learn/choosing-your-tools/paying-from-anywhere"},"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) 'The costs that are not on the invoice', in Choosing and Using the Tools. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/choosing-your-tools/hidden-costs (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":"The costs that are not on the invoice","courseTitle":"Choosing and Using the Tools","module":{"index":4,"slug":"what-it-costs","title":"What it actually costs"},"answer":null,"keyIdea":"The invoice is the small number: verification time, rework, learning and automation bias decide whether a tool pays, and the returns are largest exactly where you can check an answer faster than you could produce it.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-invoice-is-the-small-number","runs":[{"text":"The invoice is the small number"}]},{"t":"p","runs":[{"text":"A tool that costs $20 a month can easily cost you far more than $20 a month. The costs that do not appear on any invoice are real, they are often larger, and they are the reason two people with the same subscription have completely different experiences of whether it was worth it."}]},{"t":"h2","id":"verification-time","runs":[{"text":"Verification time"}]},{"t":"p","runs":[{"text":"This is the big one. If a tool produces a draft in thirty seconds that would have taken you twenty minutes, but you then spend twenty-five minutes checking it because you cannot afford to be wrong, the tool has cost you five minutes."}]},{"t":"p","runs":[{"text":"Whether verification is expensive depends on a property of the task, not of the model: "},{"text":"how long does it take to check an answer compared with producing it?","b":true}]},{"t":"ul","items":[[{"text":"Cheap to check.","b":true},{"text":" Code that either runs or does not. Arithmetic you can recompute. A translation you can read. A summary of a document in front of you. Here the tool is nearly pure gain."}],[{"text":"Expensive to check.","b":true},{"text":" A legal citation, a factual claim about a subject you do not know, a figure taken from a long report, a recommendation whose consequences appear months later. Here the verification can cost more than the work."}]]},{"t":"p","runs":[{"text":"The practical consequence: these tools give the largest genuine returns on tasks where you can verify quickly, and the largest illusory returns where you cannot. Notice which kind you are doing."}]},{"t":"h2","id":"rework-and-the-confident-wrong-answer","runs":[{"text":"Rework and the confident wrong answer"}]},{"t":"p","runs":[{"text":"A wrong answer you catch costs you the time to catch it. A wrong answer you do not catch costs whatever it costs: a supplier paid twice, a clause misread, a patient given the wrong advice, a customer told something untrue."}]},{"t":"p","runs":[{"text":"You cannot price this directly, but you can price it roughly by asking what the last mistake in this area cost, and how often it would have to happen to wipe out the time saved. For work where being wrong is expensive, that arithmetic frequently says use the tool for drafting and never for deciding."}]},{"t":"h2","id":"learning-and-switching","runs":[{"text":"Learning and switching"}]},{"t":"p","runs":[{"text":"Getting good with a tool takes ten to twenty hours spread over a few weeks. That is a real cost, paid once, and it is why switching tools every month is expensive even when each individual tool is free. You are permanently in the beginner phase."}]},{"t":"p","runs":[{"text":"It is also why the six-month review cadence is right. Frequent enough to notice a real change, infrequent enough that you keep the skill you built."}]},{"t":"h2","id":"cognitive-costs-stated-honestly","runs":[{"text":"Cognitive costs, stated honestly"}]},{"t":"p","runs":[{"text":"Two of these are documented and worth knowing about."}]},{"t":"p","runs":[{"text":"Skill maintenance.","b":true},{"text":" If you stop doing a thing yourself, you get worse at it. This matters where the skill is the point (a student learning to write, a junior developer learning to debug) and matters much less where it is not, such as formatting a spreadsheet. The evidence on this is still developing and the effect size is contested. What is not contested is the mechanism: practice maintains ability, and outsourcing practice removes it."}]},{"t":"p","runs":[{"text":"Automation bias.","b":true},{"text":" People accept a machine’s answer more readily than a colleague’s, and check it less. This is a well-established finding across decades of research into other automated systems, and there is no reason to think it does not apply here. It is worth designing around: decide in advance which categories of answer you always verify, so the decision is not made in the moment when the answer looks convincing."}]},{"t":"h2","id":"the-costs-that-go-the-other-way","runs":[{"text":"The costs that go the other way"}]},{"t":"p","runs":[{"text":"Being fair, some hidden costs are negative: the tool saves you things that never appear on an invoice either. The email you were dreading that now takes four minutes. The document in a language you read slowly. The code you did not have to ask a busy colleague about. The blank page you did not have to face. For a great many people these are the actual value, and they are just as invisible to the accountant as the verification time."}]},{"t":"h2","id":"putting-a-number-on-it","runs":[{"text":"Putting a number on it"}]},{"t":"p","runs":[{"text":"For two weeks, keep one line per day: "},{"text":"minutes saved, minutes spent checking, one mistake nearly made","i":true},{"text":". It is ninety seconds a day and it produces a real answer."}]},{"t":"p","runs":[{"text":"Most people find the tool is worth substantially more than they pay on the tasks they can check, and worth much less than they thought on the tasks they cannot. That is a useful thing to know about your own work, and no review will tell you."}]}],"check":{"question":[{"text":"Two colleagues use the same assistant. One saves several hours a week; the other finds it costs time. Their tasks differ in one property. Which?"}],"options":[{"text":[{"text":"How complex their tasks are, since the model handles simple work well and struggles with complex work."}],"correct":false,"tempting":[{"text":"Sorts tasks by difficulty when the deciding property is checkability; hard code that either runs or does not is cheap to verify."}]},{"text":[{"text":"How long their prompts are, since the one saving time has learned to give more context."}],"correct":false,"tempting":[{"text":"Credits prompt skill, which changes answer quality without changing how long verification takes."}]},{"text":[{"text":"How long checking an answer takes against producing it, which decides if verification eats the saving."}],"correct":true,"tempting":null},{"text":[{"text":"How much of the work is drafting rather than deciding, since drafting is what these tools are designed for."}],"correct":false,"tempting":[{"text":"Restates a sound guideline rather than the mechanism, and drafting a claim you cannot verify is still expensive."}]}],"order":[1,2,3,0],"why":null},"prev":{"slug":"three-billing-shapes","title":"Subscriptions, credits and pay-as-you-go","minutes":7},"next":{"slug":"paying-from-anywhere","title":"Paying for it from where you live","minutes":8}}}