Let’s talk about the thing you should have killed two years ago, and the fact that it’s still running.
You know the one. The product with eleven users. The newsletter four hundred people subscribed to and nobody opens. The partnership that made sense until the man who set it up left the company. The hire for a role that stopped existing in March.
It’s still going. It still has a weekly report. Somebody still puts it on a slide once a quarter, in the small font, down at the bottom, where the numbers go to hide.
And nobody is going to end it.
Not because it’s good. Because ending it costs more than keeping it.
Picture the meeting that would have to happen to kill it. Somebody has to sit across from the person who built it and say this isn’t working. Somebody has to admit the money is gone. The team that spent eighteen months on it has to hear that the eighteen months were a loss instead of tuition. And the person who says stop takes on a reputation, because the story will be told as “he killed it” and never as “he was right.”
So the meeting gets postponed. And postponing it is free.
That last part is new. It used to cost money to keep a bad thing alive. Rent, salaries, a licence, a desk, a machine that needed servicing. The world charged you a monthly fee for your own indecision, and that fee was the pressure that eventually forced the meeting.
Now the report writes itself. The dashboard updates itself. The follow-up email drafts itself, and it is polite, and it is clear, and it says nothing. You can keep a dead thing breathing for years and never once feel it in your bank account.
Which means the pressure is gone, and the meeting never happens.
Here is the part people miss. The machine has no opinion about whether your product should exist. None. It has an instruction to continue, and it will continue until something stops it.
Ask an agent to retry, and it retries. Ask it to extend the plan, and it extends the plan. Ask it to keep working on the thing, and it keeps working on the thing at three in the morning while you sleep, cheerfully, at a cost so small it never shows up in a conversation.
The machine will never bring you the news you need. It doesn’t get bored of the project. It doesn’t feel the embarrassment that would force the meeting. It has no stake in the ending, so it will never raise one.
That’s not a flaw in the design. Persistence is the whole product. Every agent you buy is sold on the fact that it doesn’t quit.
Now think about what that does to a person, not a company.
There’s a version of this that has nothing to do with business. The folder of unfinished writing. The course you bought and stopped. The relationship you’re technically still in because nobody said the words out loud. Deep down, all of it was held open by the same thing: the day you’d have to sit down and decide.
And now the deciding can be outsourced. The app keeps the streak. The system keeps the reminder. The assistant keeps the plan alive and sends you a gentle summary every Sunday, so the thing never fully dies and never actually lives.
An open file is not a decision. It’s a decision that has been postponed so many times it starts to look like a fact.
There’s a deeper cost, and it’s not about money or time. It’s about meaning.
A thing without an end has no shape. You can’t learn from it, because the lesson is always arriving next quarter. You can’t be proud of it, because it’s never finished enough to hold up. You can’t grieve it, because it’s not dead — it’s just still there, like a light left on in a room nobody uses.
Every story you have ever been moved by has an ending. Not because endings are sad, but because they’re what makes the middle mean anything. Take the ending away and you don’t get more life. You get a thing that goes on and on and teaches nobody anything.
And there’s one more layer to this, which is the uncomfortable one.
The record never closes. If a thing is still running, nobody has to be accountable for how it went. You can say we’re still learning, we’re still iterating, the numbers are early. That sentence buys you another year, and it is almost impossible to argue with, because the machine will manufacture the next report that makes it look reasonable.
A business full of things that never ended is a business where nobody ever has to be wrong. That sounds comfortable. It’s actually how an organisation loses the ability to tell the difference between a bet and a mistake.
So what do you actually do. Two things, and neither is complicated.
First. When you start something, put a kill date on it. Not a review date. A kill date.
A review date asks “how is it going?” and the answer is always “we’re making progress.” A kill date asks a harder question: on this date, this stops unless someone I trust stands up and argues for it. That flips the default. The thing has to earn its continuation out loud. It has to be defended in a room by a person who believes it, with a reason that isn’t a report the machine generated.
Second. Name the person who is allowed to say stop, and then cover them.
Because the reason endings don’t happen isn’t that people can’t see the truth. It’s that seeing the truth is professionally expensive. The person who calls time on a project gets remembered as the one who gave up, and nobody remembers the money they saved. So if you want endings in your company, you don’t need more data. You need one person whose job it is to ask, every quarter, “if we started this today, would we start it?” — and who does not get punished for answering honestly.
That question is the whole thing, by the way. Not “is this working.” Everyone can argue about whether something is working. “Would we start it today” has one clean answer, and the machine can’t help you dodge it.
Here’s what I keep coming back to.
The machine can keep anything alive. Forever. Cheaply. That’s a genuine gift for the things that deserve it.
But it will also keep alive every dead thing you were too tired to end. And it will do it politely, and it will call it momentum, and it will hand you the report that proves it’s fine.
Endings are the one part of this that can’t be automated. Somebody has to sit in a room and say it’s over, and take the cost of saying it. That cost is the price of a life that has a shape.
The machine will never tell you it’s over.
You’re the only one who can. And that’s not a burden. It’s the last thing in the whole system that’s still yours.