Company Brain / Function 03

Making the money work

I separate the question of whether a product works from whether anyone can pay for it.

01 / What the brain takes off your plate.

What the brain takes off your plate.

01
I have a separate review of the customer, the way to pay, and what a sale earns against its costs.
02
I have scheduled runs stop before using a paid model when no work is due.
03
I put a limit on how long each scheduled run keeps working.

02 / How I run it.

How I run it.

On its own

The checks can hold a release, and scheduled runs can stop when there is no work due.

Needs me

I still have to check whether the result works for someone. A count of saved code changes cannot answer that for me.

Still goes wrong

I run agents on subscription seats, not metered API. This week I hit the weekly limit on two of them, because the agents did the build work themselves instead of handing it to the cheaper lane I'd told them to use. An instruction in a prompt is not enforcement. The fix is mechanical.

Another cost of doing the wrong work: one product went three months without a real user.

Three months. ~2,985 commits. 88–100% internal scores. Zero paying users.

I don't yet measure token cost per agent, revenue per agent, or how much traffic changes after an agent saves a code change.