The Brief · The Durability Curve

What Did They Have to Check or Redo?

Wed, 14 Oct 2026 00:00:00 GMT

An AI-drafted update, summary or email can cost the person who gets it time you do not see. Ask them one question, then change one thing.

Your AI drafted the weekly update in three minutes. You read it through, it looked right, and you sent it.

What happens next happens on someone else’s desk. They may trace a figure back to its source, put back the context the summary dropped, or read it twice to find what you need from them. None of that counts as time you spent, so none of it counts against the time AI saved you.

The sender sees the saving, and the recipient sees the repair.

Researchers at BetterUp Labs and Stanford’s Social Media Lab have a name for work like this: workslop, AI-generated work that looks finished but lacks the substance to move the task forward. In their 2025 survey of 1,150 US desk workers, 40 per cent said they had received it in the past month. Those who had received it put their own estimate at nearly two hours per incident, and half of them said they now saw the sender as less capable. Thirty-two per cent were less likely to want to work with that person again.

So ask the person who received yours.

Pick one handoff you send every week: a status update, a research summary, a client email, a meeting recap. Send it the way you normally would. Once the recipient has used it, ask:

What did you have to check or redo?

The exact words matter. Ask whether it was useful and you invite a polite yes. Ask how much time it saved and you are back on your own side of the handoff. Ask what they had to check or redo, and the honest answer is a list, or “nothing”.

Take the first thing they name and fix it next time. If they reopened your sources, put the source beside each claim. If they hunted for the request, put it in the first line. Then ask again after the next handoff. You want that list to get shorter each time, until there is nothing on it.

The output does not have to be wrong for this to matter. A correct summary can still leave out what the decision turns on. A well-written email can still bury the request. A polished update can still leave nobody sure who owns the next step. The person who received it can tell you which of those happened.

For the wider question of turning faster output into trustworthy work, read AI Made You Faster. It Did Not Make You Safer.

Your AI Complaint Says Things You Never Said

Mon, 05 Oct 2026 00:00:00 GMT

A study found AI-written complaints win relief more often. Its own AI edit of one says ‘I do not own a cell phone’. The consumer never said so.

You have a complaint about your bank, and it reads the way you feel. Asking a chatbot to make it clearer is a sensible move. A study of 1,134,512 complaints to the US Consumer Financial Protection Bureau estimated that using AI raised a complaint’s chance of winning relief from the firm by about 7 percentage points, where relief can be money or another fix. The paper says the AI did this “without altering factual content.”

What the study’s own edits say

To test why, the researchers ran experiments. People imagined handling complaints and said how likely they would be to pay out. Some complaints were as consumers wrote them, and some had been edited by ChatGPT. The researchers published the complaints they used.

One consumer had tried to get a credit report online. They wrote that TransUnion already had their landline, and that the form requested a cell phone number, “which I did not fill in.”

The study had ChatGPT edit that complaint three times, to be more clear, more coherent and more professional. All three versions say the consumer doesn’t own a cell phone. The professional one reads: “Although I had already provided my landline number, I could not fill the cell phone field as I do not own a cell phone.”

It may be true. The consumer never said it, and these edits were among the complaints people rated. That doesn’t overturn the 7-point finding, which comes from complaints people actually filed. It does mean a chatbot can fill your complaint’s gaps with the likeliest story, and that story goes out under your name.

Three steps

Write your facts first. A short numbered list: what happened, when, what the firm said or did, what it cost you, and the one thing you want them to do. Don’t skip that last item: the prompt below asks the chatbot to end with how you want the complaint resolved, and if you don’t say, it will guess. Leave out account numbers and any health or banking details, and add them yourself afterwards.
Use an ombudsman’s own prompt. The Public Service Ombudsman, which handles complaints about the NHS in England, publishes this one, and nothing in it is specific to the NHS. Paste it, then your list:

Rewrite the text below as a clear, polite complaint. Use only the information I have provided. Do not add or invent new facts. Be concise and professional, but not too formal. Include a closing paragraph suggesting how I would like the complaint to be resolved.

Read it against your list. One sentence at a time. Anything you never said comes out. Any law or rule you didn’t write yourself stays out unless you can confirm it on the regulator’s or ombudsman’s own website.

The third step is the one that matters. We gave the ombudsman’s prompt to Claude Sonnet 5 and GPT-5.6 with three of the study’s real complaints. With the cell-phone complaint, Claude still wrote, “I do not have a cell phone number to provide.” With a debt-collection complaint, it called the consumer’s $1,000 claim “statutory damages”, a legal term the consumer never used. GPT stayed closer to what the consumers wrote, though one of its three rewrites still implied the consumer had raised concerns before. Reading each version against the original catches all of these. That was three complaints and one run per model, read by an AI agent rather than a person, so take it as an example, not a rate.

Why it works

A chatbot writes the most plausible complaint it can, and only you know which parts happened. The ombudsmen put the check on you. The Public Service Ombudsman says, “Always check AI-generated text to make sure it’s accurate and reflects your own personal experience.” The Financial Ombudsman Service says, “You are responsible for making sure the resulting text is accurate, relevant and truly represents your experience. If you don’t, you may experience delays.”

Let the AI do the clarity. The facts are yours.

Where it goes next

In the UK, a bank or other financial firm has up to 8 weeks to answer most complaints. For fraud and scams, payment services or electronic money, it usually has 15 working days. If it hasn’t replied in time, you can take the complaint to the Financial Ombudsman Service. If it has sent a final response you’re unhappy with, you have 6 months from that response. The service is free, and you don’t need a lawyer.

In the US, before you complain to the CFPB about a credit bureau over an error on your report, the law requires you to dispute it with the bureau first. Then file with the CFPB, which sends your complaint to the company. You generally can’t file a second complaint about the same problem, so put everything in the first one.

An earlier piece, Show It an Example, showed the AI doing this with tone of voice: whatever you don’t pin down, it fills with the likeliest version.

One Question for Any AI Model Headline: Can You Use It?

Fri, 25 Sep 2026 00:00:00 GMT

It may be in a mode you didn’t open, on a plan you don’t have, or kept for partners. Here’s the check.

OpenAI has announced GPT-6 Astra. Say you read the headline, opened ChatGPT and went looking for it. On one Plus account we checked on 25 September, the model menu in Chat mode offered GPT-5.6 Sol and GPT-5.5, two clicks in. Astra wasn’t on the list. It was there, named on the menu once we switched to Work mode, and OpenAI’s help page says as much: “Plus plans include GPT‑6 Astra in ChatGPT Work and Codex.”

That is the small version of a wider gap. The model in the headline can be somewhere in your app you didn’t look, or not on your plan at all, and the headline won’t say which.

Launches reach customers in levels

Each lab says who gets the model further down its own page. OpenAI says Astra “will become available to all ChatGPT Plus, Pro, Business, and Enterprise users”, and Free isn’t listed. Google says Gemini 3.8 Flash “is available to Google AI Pro and Ultra subscribers” in the Gemini app. xAI says Grok 4.7 “is available today in Cursor and Grok Build” and through its API, and on one account we checked, every mode in the Grok app pointed to Grok 4.6.

Each also keeps something for a smaller group: GPT‑6 Pro, powered by Astra, for OpenAI’s Pro, Business and Enterprise plans, a Cyber version of 3.8 Flash for Google’s “trusted defenders”, and “invite-only access” to Grok 4.7’s red-team capabilities for select xAI partners.

Find it on your plan, in every mode

So before you repeat a model headline, act on it or pay for it, ask one question: can I use it? There are three answers: yours, yours at a price, or not yours.

To find out, open your app and look for the model on your plan, in every mode the app has. If the menu shows modes such as “Fast” or “Expert”, hover over each one, because some apps name the model only there. If it still isn’t there, search the model’s name plus “available” and read that sentence on the lab’s newest page, which may be a help page rather than the launch post.

One Anthropic model family shows all three answers. Anthropic says Claude Fable 5.1 and Claude Mythos 5.1 “are the same model, but with different levels of safeguards”. Its help pages say that, as of September 2026, Fable 5.1 is “included as a standard part of your plan” on Max, where you can spend “up to 50% of your weekly usage limits” on it at no extra cost: yours. On Pro, it runs “on pay-as-you-go usage credits”: yours at a price. Mythos 5.1, which Anthropic says has “the strongest cyber capabilities of any model we’ve released”, is “available only through our trusted access programs”: not yours, unless you are in one.

The page says what you should get

In 2023 the researcher Irene Solaiman described model releases as a gradient from fully closed to fully open. We’d add that plans make a smaller gradient inside the apps we use. Launch posts also go stale: Anthropic’s June post said Fable 5 would be included on Pro through 22 June, and its help pages now carry the current terms. So read the lab’s newest page to find out what your plan should get, then check your app to see what has actually reached you. For the frontier launches since 12 August, The Access Gap keeps that record: when each model reached the API, the app, the cheapest paid plan and the free tier, with the lab’s own page behind every entry.

A model in the headline isn’t yours until you find it on your plan, at your price.

Even once you find it, the model’s name is only half of what you get, and I wrote about the other half in Same Model, Different Product.

Show It an Example

Thu, 17 Sep 2026 00:00:00 GMT

Why “make it distinctive” gets you the least distinctive thing there is, and what to paste instead.

You wanted it to write something with a particular feel. A tagline, a bio, an email in the right tone, a name that sounds like you and not like everyone. So you were careful about it. You asked for punchy. Modern. Distinctive. You told it to steer well clear of the corporate stuff.

And it came back polished and completely generic. The kind of line you have read a hundred times on a hundred websites. You asked for distinctive and got the opposite, so you added more words, and got a longer version of the same thing.

The words were the problem. Being more specific in adjectives is not being more specific. “Distinctive” is not a target, it is a direction, and there is an enormous crowd of distinctive-ish things it could point at. Told to head in that direction, the machine has to decide what you mean, and it often lands somewhere familiar in that crowd: polished, broadly compatible with the word you gave it. That is where generic comes from. You named a direction without showing it where along that direction you wanted to go.

I watched this happen. I asked for five taglines for a small bookshop, and said make them charming, memorable and distinctive. It gave me the flowery, template version, the exact voice you would guess: “trade the noise of the world for the turn of a page.” Asked to be distinctive, it produced the least distinctive thing on offer.

Paste an example, and say match this

So stop describing the style, and show it one.

Find something that already has the feel you want. A tagline you wish you had written, a paragraph whose voice you would happily borrow, an email whose tone you admire (keep out anything you would not want to put into the tool you are using). One is enough to start, and a few can work better. Paste it in and say: match this. That is the whole move, and it is shorter than the paragraph of adjectives you were about to write.

When I dropped the adjectives and instead pasted three taglines I liked, plain and dry and lower case, and said only match these, it answered in the same plain voice. Not by repeating the three I gave it, but with new ones in the same plain, lower-case voice: “books you’ll actually read.” “fewer books, better ones.” “just books, really.” Same machine, same bookshop. What I had taken away was the adjectives; what I had added was something to aim at.

The example does not have to be from the same world. A tagline you like from a coffee brand can teach it the voice for your bookshop. Think of it as the target to aim at, and you can still keep, edit or throw out whatever comes back.

Why an example beats a paragraph

An example is a thing it can see and follow. An adjective is only a word, and a word covers everything the word could mean. “Warm” spans a birthday card and a hostage negotiation. Show it one line that is warm in the way you mean, and suddenly “warm” has an address.

That is the difference between pointing at a target and naming a direction. It can hit a target you show it. Give it only a direction and it often lands somewhere familiar.

An example is a target it can see and match. An adjective is only a direction, and it fills a direction with the ordinary. Show it the thing; do not describe the thing.

That is also why this holds up as the tools change. It is not a trick tied to today’s model. However good the next one is, it cannot see the specific thing in your head, and a description will rarely carry it across as cleanly as one real example. The surest way to hand over your specific thing is to show it an example of it.

Where it catches you out

It will follow your example closely, and that includes a bad one. Grab the first thing to hand because it was convenient, and you will get a clean, confident version of something you did not actually want, which is harder to spot than an obviously generic miss.

So before you paste it, read the example back and ask whether it is really the thing you want, not just the thing you found. The move carries your taste across to the machine. It does not give you taste, and it will follow a lazy choice as readily as a good one. When the thing you are trying to get right is bigger than a tagline, the same rule holds: a better model still needs to know what your version of good looks like. I wrote about why the setup you give a model beats the model itself for anyone whose version of this has moved past writing taglines.

Ask for the Step Before the Answer

Wed, 16 Sep 2026 00:00:00 GMT

Before you trust an AI answer you cannot check, ask for the step it was built from.

You asked it to do something finished. Summarise a forty-page report. Pull a growth figure out of a spreadsheet. Read five documents and tell you which supplier to choose.

What came back was clean, quick, and shaped like competent work. So you sent it on, quoted the figure, went with the supplier. It looked right, and looking right was the whole of what you had.

A finished answer is where good work and bad work can start to look most alike. Both arrive fluent, confident and complete. And the finished answer is usually the most expensive thing to check: to test the summary you would have to read the report, to test the figure you would have to rebuild the sum, to test the recommendation you would have to read all five documents yourself. Checking the thing you got back costs you the very work you were trying to hand off.

A right answer and a wrong one arrive under the same smooth surface, and the surface is the most expensive part to check.

Ask for the step before the answer

So do not run your check there. Ask it to stop one step earlier, at something cheaper to check, and check that first.

Before it writes the essay, ask for the outline: one line per point, in the order it will make them. Before it gives you the number, ask for the inputs, the assumptions, and the sum. Before it summarises the report, ask for the passages carrying its main claims, with page numbers. Read that. Only when the step underneath holds up do you let it finish the job.

Each is a smaller part of the task. A summary stands on specific claims and passages. A number stands on inputs, assumptions and a formula. An essay stands on a structure. The finished answer has compressed the sources, assumptions and decisions into one result. Moving one step back exposes those parts separately, where they are easier to inspect. A mistake that could have taken an afternoon to trace through the finished work may be sitting in the open a few lines earlier.

You will not always catch it, and some errors still need an expert eye. But you have made the error cheaper to catch, often much cheaper.

Move your check to the cheaper step: the outline, the inputs, the sources. Approve that, then let the finished answer be built from it.

Do not start your check where the work ends. Ask for the step underneath it, while the work is still cheap to inspect.

The way it fools you

There is a version of this that feels the same and is not. You let it answer, then you ask how it got there.

What comes back is an explanation generated after the answer already exists. You cannot know whether it faithfully represents how the answer was produced, and it is too late either way: you are inspecting an account of the work instead of creating a checkpoint before it.

Asked first, the same request does something else. The answer does not exist yet, so what you approve becomes the thing you can require the answer to be built from. That is the whole difference. Ask afterwards and you get an explanation. Ask beforehand and you get a constraint.

Asked first, the step becomes a checkpoint for the answer that follows. Asked afterwards, the explanation is generated with the finished answer already in place.

The two messages

The habit is two messages. Before it starts, send the first:

“Before you give me the final answer, show me the part I should check first. For this task, that might be the sources, the inputs and assumptions, or the outline. Stop there and wait for me.”

Then, once you have read that and it holds up:

“Now write the final answer using only what I approved.”

The first message moves your check to the cheap step. The second reduces the chance that the finished answer drifts away from the step you signed off.

That is the whole thing. Ask for the step before the finished answer. Check it first. Then make the final answer come from what you approved.

This is the cheap version of a larger problem: AI is most tempting to trust where checking it is hardest. I wrote about why an AI looks best exactly where you can check it least for the cases where getting this wrong is expensive.