Clean the Fishbowl, Don't Treat the Fish

At a leadership offsite recently, one line outlasted every slide: clean the fishbowl, don't try to treat the fish.

The picture is simple. If the fish in a bowl are sick, you can medicate each fish, one at a time, forever. Or you can clean the water. If the water is the problem, only one of those works.

I wrote the line down and then, underneath it, a second one: easy to remember, easy to misuse if I forget it.

The first line is not mine. I heard it from a woman at that offsite, and the credit is hers.

This essay is about both halves.

Why the metaphor is right

W. Edwards Deming, writing about quality in Out of the Crisis, estimated that 94% of troubles and possibilities for improvement belong to the system, and only 6% to special causes, including individual people. You do not have to accept his exact number to recognise the pattern. When the same problem keeps happening with different people in the seat, the seat is the problem.

There is a second cost to treating the fish. People in a bad bowl stop telling you what is wrong with the water, because every conversation turns into a conversation about them. The problem goes quiet, not away. Psychological safety is not a nice extra here. It is how you find out what the water is like.

What a bowl looks like

Here is an illustrative case, not a real one, but it will be familiar to anyone who has run an operation.

A depot keeps missing its afternoon dispatch cut-off. Not every day, but three or four times a fortnight. The easy conclusion is that the afternoon team needs coaching on urgency.

Look at the bowl instead. The final load list reaches the team twenty minutes before cut-off because it comes from a system that batches updates. The screen that shows it is at the far end of the dock. And the person who used to chase late items was moved to another shift during a roster change.

None of those things is about the team's attitude. Each is a wall of the bowl. Coach the fish and nothing changes, except that good people start to feel blamed.

I am as guilty of this as anyone. I call myself a fixer, and as a fixer my instinct in a depot like that would be to fix the afternoon team, or fix the late list. Most of the time the real job is fixing the environment so people can do their best work. When I skip that, it has a diminishing effect on people, and I do not always notice it happening.

Agents live in bowls too

The same thing happens with AI. When an agent keeps getting something wrong, the instinct is to write a sterner prompt. Try harder. Be more careful. Do not do that again.

I have done exactly this. My morning brief started quietly dropping sections after I moved it to a new pipeline. My first move was the fish move: tell the agent, in firmer words, to include every section. It still dropped them some days.

The fix was the bowl. I wrote down the list of sections the brief must contain, and added a check in code that compares the brief against that list and puts a warning at the top when something is missing. Now a missing section is visible at 6am instead of noticed a week later. I did not make the agent more careful. I made carelessness impossible to miss.

The same thing happened with my private notes. A rule in a file the tools never read did not stop two of them copying a private passage. A guard in code did. I wrote that one up in Prose Is Not a Control.

Prompting harder is treating the fish. Fixing the workflow is cleaning the bowl.

Easy to misuse

Now the other half. The fishbowl can become an excuse for anything. Every miss is the system. Nobody is accountable. That is not what it means.

Standards still apply. Some problems really are about one person, and pretending otherwise is unfair to everyone else in the bowl. The metaphor is not a verdict. It is an order of operations: look at the water first.

One wall, two weeks, then coach

The discipline I am trying to follow is this.

Change one wall. Pick the one thing about the system most likely to be causing the problem. Change only that, so you can tell whether it worked. Tell the team what you changed and why.

Watch for two weeks. Long enough to see a pattern, short enough that nobody forgets the change. Write down what moved.

Then coach. If the problem has gone, you have fixed it for everyone who will ever sit in that seat. If it has not, you have earned the right to a conversation about the person, and you can have it without guessing.

This is harder than coaching. Coaching is one conversation. Changing a wall usually means talking to another team, a system owner or a roster. That is exactly why it is worth doing.

The question I am trying to ask before every hard conversation: what would a competent, well-meaning person do in the bowl I built?