The Cheap Ones Erode

The commitment most at risk in your company is the one an engineer could undo before lunch

Pick something your company holds constant that one person could reverse in an afternoon: a rule about what an integration is allowed to read, or a default that gets set the same way every time. That is the one at risk, and it will not feel like it.

Nothing below is measured. It is drawn from reading scaling companies closely, it is offered as observation, and the argument is checkable inside your own company in about ten minutes, which is what the exercise at the end is for.

Raising something costs more than the thing is worth

Escalation is not free. Putting a change in front of a group costs a meeting, some goodwill, and a small deposit into a reputation for obstruction that a person spends the rest of the year carrying. Those costs are roughly fixed. They do not scale down when the thing being raised is small.

So when somebody proposes reversing a commitment that would take an afternoon to undo, the arithmetic runs against raising it. The reviewer would spend more than the decision is worth, and everybody in the room can see that, including the person who wanted to object. The change goes through. It has nothing to do with whether anyone cares.

Do that several times over a year or two and the commitment is gone, with a defensible reason attached to every step and no meeting anywhere in the record.

What erosion looks like from inside, where it looks like competence

The integration was allowed two fields, and a partner needed a third. The case for it was good and the scope widened for that partner alone. A quarter later a second partner asked, and refusing would have looked arbitrary. Then somebody generalised the permission, because handling each special case individually had become the slowest part of onboarding, and generalising it was a real improvement.

Each of those calls was reasonable on the information in front of the person making it, and each was smaller than anything that gets escalated. Nobody proposed opening the integration up, so nobody rejected it, and it is open.

This is why the commitment does not appear in a post-incident review as a decision. There is no decision to find.

The blast-radius rule is right, and it is a rule about decisions

Most engineering organisations govern by blast radius, and they should. Sorting by what happens if this turns out to be wrong is the cheapest way to put scarce reviewing attention where it changes outcomes, and it beats sorting by budget, by seniority or by who asked.

The rule was built for choices not yet made, and for those it is close to unimprovable. What it reads is how long the change would take to undo, and for a commitment the company already depends on that figure has come loose from what losing the thing would cost. When those two numbers should part company is a separate argument and it is made elsewhere in this series. The narrow point here is that while they are apart, the rule keeps reporting the smaller one.

The property that protects it is the same one that hides it

Being cheap to reverse is what makes a commitment low-risk to change. It is also what keeps every individual change below the threshold that any review process uses, and review processes are organised around cost. A commitment this cheap is invisible to them by construction rather than by oversight.

Expensive commitments do not need a list. Their own cost defends them: nobody quietly migrates a platform on a Thursday. The list exists for the ones with nothing else protecting them.

What to do before your next planning session

Take ten minutes and pick the commitment you would be least worried about losing. That instinct is the signal, so use it rather than arguing with it.

Then ask one question about it: if somebody removed this next month, whose approval would they need? Where the honest answer is that removal would not need approving by anybody, you have found the exposed one.

That is the output, and the exercise stops there. Put the item on the agenda for your next planning session as one line saying this is the one at risk, with a date beside it for when somebody will check whether it is still intact. A flag and a date.

What to do about the item you flagged is a different question with several answers, and picking one before you know which commitment you are protecting is the wrong order.

The platform and the contract can look after themselves. Which of the small things your company holds constant would go this quarter, quietly, with a good reason attached to every step?