The Delicate Art of Uncommitting, Revisited: How I Use AI to Decide What Gets a Yes

In 2015 I wrote a short piece for this site called The Delicate Art of Uncommitting. It was an excerpt from a book on getting organised, and it offered five scripts for backing out of things graciously. I read it again this week for the first time in years. The scripts still hold up. The woman who wrote them, however, was solving the wrong problem. She had become very good at getting out of commitments and had not yet learned how to stop making them.

Overcommitting is my oldest flaw. I have run companies, raised four children, trained on some of the largest stages in the world, and I still catch myself saying yes to a podcast, a panel, a board, a favour, a “quick call” that is never quick. So this year I did something I would not have predicted a decade ago. I built a filter, gave it to an AI, and asked it to hold me to it. This is the story of what happened, why the science says most of us need something like it, and the exact prompt so you can run it yourself.

Why smart people keep saying yes

It is tempting to treat overcommitment as a character issue. The research says it is closer to a measurement error. In a foundational study, Gal Zauberman and John Lynch found that people expect far more “slack” in their future time than in their future money. We believe next month will be roomier than this month, so we book it up. The 2005 paper in the Journal of Experimental Psychology: General showed that this expectation of future slack “is more pronounced for time than for money,” which is precisely why the same person who would never spend money she does not have will cheerfully spend hours she does not have.

Behavioural economists have a name for what follows. They call it the “yes… damn” effect. You say yes in March to something in June, and in June you say damn. Harvard Business School’s Ashley Whillans puts the corrective bluntly: the best predictor of how busy you will be next week is how busy you are right now. Not how busy you hope to be. How busy you are.

We would never spend money we do not have. We spend hours we do not have every single week.

The second error is about other people. We overestimate how badly they will take a no. In a 2024 series of five studies in the Journal of Personality and Social Psychology, Julian Givi and Colleen Kirk found that people who declined invitations “significantly overestimated how negatively the inviter would react,” assumed the host would fixate on the refusal rather than the reasoning behind it, and exaggerated the damage to the relationship. The hosts, it turned out, were far more understanding than the decliners feared.

And the third error explains why the requests keep coming. Cornell’s Vanessa Bohns has spent more than a decade studying what happens when one person asks another for something. Across studies involving more than 14,000 people, askers underestimated how often they would get a yes by roughly half. The reason is not that people are generous. It is that refusing someone to their face is uncomfortable, and we will do a great deal to avoid that discomfort. If you are a leader, a founder, or simply the reliable one in your circle, you are being asked more than you realise, and you are saying yes more than you should.

The cost is not abstract. A May 2026 survey of 1,000 U.S. small business owners found that 84 percent had sacrificed their health, relationships, or mental wellbeing for the business, more than half lose sleep over it at least a few nights a week, and only 22 percent would describe their mental health as thriving. Very few of those owners are short on discipline. Most of them are long on yes.

The four words

Years ago I settled on a simple test for any new commitment. It has to bring joy, impact, profit, and peace. Not one of them. All four, or a very honest three with a reason. Joy is whether I actually want to do it, not whether I want to have done it. Impact is whether it changes something for someone. Profit is whether it moves the business, the mission, or the family forward in a way I can name. Peace is whether I can say yes and sleep. If I have to talk myself into any one of the four, the answer is already no.

The test is good. My follow through was not. Filters only work if you run them, and in the moment of being asked, with a kind person waiting and Bohns’ discomfort doing its work, I did not run it. I said yes and ran it later, which is the “yes… damn” effect with extra steps.

Where the AI came in

I use Claude as a working assistant. I have named it Pembrooke, because it helps to talk to something with a name, and in its standing instructions I wrote one line that has changed more about my calendar than any productivity system I have ever tried: my biggest flaw is overcommitting, and you will make sure I do not say yes to anything that does not bring joy, impact, profit, and peace.

Here is what that looks like in practice. This week I floated an idea for a new blog series, the one you are reading. I was enthusiastic. I expected a list of titles. What I got first was a question about whether this was a weekly obligation on top of a book I am writing, a company I am running, and a webinar I am giving, and a plain observation that if it was, it would fail the peace test. I had to stop and answer honestly. It was not an obligation. It was thought leadership, something I would write when I had something to show. That reframing is the only reason the series exists in a form I can sustain.

The AI did not tell me no. It asked me which of the four words I was pretending about.

The filter now runs in more than one place. James, the AI agent that handles my inbound email, carries the same four words. Not long ago a request arrived asking me to speak in front of what was described as 2,000 women. There was no fee. I would be expected to promote the event to my own audience, and in return I could offer a lead magnet. Ten years ago I would have said yes before I finished reading, because 2,000 women is exactly the room I want to be in. James was a hard no, and when I made it explain itself, the reasoning was cleaner than mine would have been. Joy, possibly. Impact, unverified; “allegedly 2,000” is not 2,000. Profit, negative, because I would be spending my list to build theirs. Peace, no, because I would have resented every hour of preparation. Three of four failed. The decline went out that afternoon, graciously, and I felt lighter reading it than I would have felt saying yes.

That is the tie between this story and the work I do for a living. As a female AI keynote speaker, most of what lands in my inbox is an invitation to stand on a stage, and the stages I say yes to are the ones where the audience is real, the organiser has skin in the game, and I can walk off knowing the room got something it could use on Monday. The four words are not a wall around my calendar. They are how I make sure that when I do say yes, I am all the way in.

I am not alone in using AI this way. Elon University’s Imagining the Digital Future Center reported in September 2026 that 27 percent of American adults now have significant personal or social interactions with AI, and among those users 53 percent seek interpersonal guidance, 39 percent have talked through career decisions, and 39 percent admit they tell the AI things they would not tell other people. Meanwhile Pew finds half of Americans are more concerned than excited about AI. I understand both numbers. I would simply point out that a tool which helps you keep a promise to yourself is the least frightening use of this technology I can think of.

The prompt

You do not need a named assistant or a custom setup. Paste the following into whichever AI you use, replace my four words with yours if you have better ones, and then describe the request you are weighing. The value is in being interrogated before you answer the human, not after.

A note on tools: I run this in Claude, where the instruction lives permanently in my assistant’s settings so I never have to remember to invoke it. It works just as well pasted fresh into ChatGPT or Gemini each time. The permanence is the point, though. If you have to remember to run the filter, you are back to relying on willpower in the exact moment willpower is weakest.

What did not work

Two things, and they matter. First, the filter is only as honest as the person answering the questions. The AI cannot feel the yes in my body, and early on I caught myself answering the joy question the way I wanted it to be true rather than the way it was. The fix was to add “including what it would cost in hours and when” to the prompt, because I can lie about joy but I cannot lie about a Tuesday.

Second, without the four words the AI defaults to helping me do more. That is what these tools are built for. Ask an assistant to help you fit something in and it will find a way to fit it in. The filter has to be given, explicitly, as a constraint. The intelligence is in the machine. The values have to be yours.

Five ways to uncommit graciously, refreshed

The filter stops new yeses. It does not undo old ones, and most of us are carrying a few. The scripts from the 2015 piece still work; I have updated them for the way we actually communicate now.

Find a replacement

“You know how I said I would help on the fundraising committee? A project has taken far more of my time than I anticipated, and I do not want to leave you in a bind. I have found someone excellent to take my place and have already spoken to her.” Bring the replacement, not the problem.

Be honest about your energy

“When I accepted, I had the capacity for it. Right now I do not, and I would rather tell you now than show up as half of myself. Please invite me next time.” Givi and Kirk’s research says the host will focus on your reasoning, not your refusal. Give them reasoning.

Trade rather than cancel

“Something landed on my desk this week that needs my full attention. If I take the next two weeks of carpooling, would you take this one?” A trade preserves the relationship because it preserves the fairness.

Name the mismatch

“Your work sounds genuinely valuable. After careful thought it does not match where my giving is focused this year. If you expand into that area, would you come back to me?” This is the profit and impact test, spoken out loud, and most people respect it.

Offer a better version

“I promised to come in early and stay late for the next two weeks. My commute is an hour and a half each way, which is almost a full working day every week. What if I work from home one day a week and put those hours into the project instead?” A better version is not a retreat. It is a redesign.

The real lesson

Twenty years ago I thought the delicate art was in the uncommitting. It is not. The delicate art is in the pause before the yes, and the honest answer to four small words. What surprised me is that it took a machine to make me keep a rule I wrote for myself. Perhaps that is the most human thing about this whole story. We have always needed a witness to our own commitments. Now some of us have one that never forgets what we said we wanted.

The intelligence is in the machine. The values have to be yours.

Run the filter on the next thing someone asks of you. Then run it on the three things already on your calendar that made you sigh when you read them. You already know which three.

If you are planning an event and wondering whether it would pass my four words, the quickest way to find out is to check my availability. I promise the answer, either way, will be gracious.