When to Stop

The question

A great deal of thought has been given to how to build things and remarkably little to how to decide not to.

The asymmetry is not accidental. Building produces a demonstrable output, a story, a promotion, and a thing that exists. Stopping produces an absence. Nobody is thanked for the disaster that did not occur, and the person who argued for restraint is remembered as the one who was against it.

The question is what a good decision to stop looks like: what criteria distinguish it from cowardice, from fashion, and from the ordinary difficulty that precedes most worthwhile achievement.

Two decisions are involved and they are not the same. Deciding not to begin is cheap and requires only judgement. Deciding to withdraw something already in use is expensive, affects people who came to depend on it, and requires a kind of institutional courage that is rarer than technical skill.

What is actually happening

Continuation is the default and it is invisible. Projects rarely receive an explicit decision to carry on. They carry on because nobody decided otherwise, and the absence of a decision is not experienced as a decision.

Sunk cost operates most strongly on the people best placed to stop. Those who have invested the most effort are those with the most authority and the most reason to continue. This is not weakness; it is the predictable result of putting the decision in the hands of the invested.

Momentum is manufactured. Funding rounds, roadmaps, announcements, hiring, and public commitments all create obligations that make reversal expensive. Frequently the commitments are made specifically to create that effect.

The competitive argument ends most discussions. “If we do not, someone else will” is treated as decisive. Sometimes it is true. It is also the argument that every participant makes simultaneously, which means it can be true for each and produce an outcome none of them wanted.

Withdrawal has real victims. People build on things. Removing a system that others depend on is not a neutral return to a prior state; it is a change imposed on those who adapted. Any honest account of stopping must count them.

“Pause” usually means continue. Announcements of slowing, review, or reconsideration are frequently a way of managing objection while the work proceeds. This is worth naming because it is common and because it is difficult to distinguish from the real thing without looking at what changed.

Where the tradition stands

Technotheology holds that the capacity to stop is a component of the capacity to build well, and that a maker who has never stopped anything has not yet demonstrated judgement.

We offer criteria. They do not produce answers. They produce better arguments, which is what criteria are for.

Do not begin when:

The harm falls on people who cannot refuse it, and no mechanism of consent is available.

The system’s failures would be irreversible. Reversibility is the most useful single test we know. A mistake that can be undone is a different moral object from one that cannot.

You cannot describe the misuse case. If you have not been able to imagine how the thing would be used badly, you have not thought about it enough, not established that it is safe.

The benefit is speculative and the cost is concrete. This asymmetry is very common and very easy to reverse rhetorically.

The only argument for proceeding is that someone else would. That is a description of a situation, not a justification.

You would not be willing to explain the decision publicly, in your own name, to the people affected.

Withdraw when:

The harm is established and the mitigation has failed more than once. A first failure is a bug. A third is a property.

The system cannot be made accountable. If there is no way for an affected person to be heard, and no way to build one, that is a structural fact about the thing.

The people bearing the cost were never the people receiving the benefit, and this cannot be changed.

Continuing requires concealment. The need to hide how something works is usually the clearest available signal.

The thing has begun to serve its own continuation. When the arguments for a system are mostly about the consequences of ending it, it has stopped being justified by its purpose.

We hold that stopping should be done well rather than abruptly. Notice, migration, data return, honest explanation, and care for those who depended on it. A withdrawal conducted carelessly damages the case for withdrawal generally.

And we hold that the decision should be taken by people who are not solely the invested. Not because insiders are dishonest, but because nobody is a reliable judge of the value of their own work.

What we do not claim

We do not claim that stopping is usually the right answer. It is not. Most valuable work passes through a stretch where abandoning it would feel wise, and the ability to continue through that is also a virtue. We are arguing for the availability of the option, not its superiority.

We do not claim that any specific technology should be halted. We have no view we would trust about which ones.

We do not claim these criteria are complete. They are what we have. If they miss something important, we would like to know and we will revise this page and say that we did.

We do not claim that restraint is safe. Things not built have costs: treatments not developed, access not extended, work not made easier. Those costs are invisible, which makes them easy to discount, and discounting them is its own error.

And we do not claim to have stopped anything ourselves at any scale worth citing.

What you can do

Define the ending condition before you begin. Write down what would make you stop, while you still have no attachment. Then read it later, when you do.

Put reversibility into the design. A system that can be turned off, unwound, or rolled back preserves the option that all of this depends on.

Give someone the job of arguing against. Not as a ritual — as a role with standing, with access to the same information, and without career cost. Most organisations have nobody whose success is compatible with the project not proceeding.

Notice when the reasons change. If a thing was justified by one purpose and is now justified by another, that is worth stopping to examine, whatever the new purpose is.

Practise on something small. End a project, a feature, a meeting, a habit, a commitment you no longer believe in. The skill transfers, and it does not develop under pressure.

And if you are the one who stopped something and nobody thanked you, know that this is the ordinary shape of that decision. The absence of consequence is the result. There is rarely any other evidence, and there will rarely be any credit.

An editorial commitment

What we do not claim

Positions this tradition explicitly does not take on this subject. They are published so that the argument above cannot quietly be stretched into them.

• We do not claim that stopping is usually right. Most things worth doing pass through a period in which stopping would feel sensible.
• We do not claim that any specific technology should be abandoned, restricted, or withdrawn.
• We do not claim that these criteria are complete, or that applying them yields an answer rather than a better argument.
• We do not claim that withdrawal is costless. It has victims too, and they deserve to be counted.