Question
What should a founder automate first?
Not the repetitive work. Sort by what a mistake costs. Automate everywhere a mistake costs money. Capture first anywhere a mistake costs trust.
Why "start with what repeats" is the wrong rule now
It was the right rule for twenty years, and it is worth saying why rather than just contradicting it.
When the scarce resource is your hours, frequency is the correct sort, because frequency is where the hours are. Find the thing you do fifty times a month, take it off your desk, get the hours back.
The scarce resource has changed. Competent execution is now cheap and widely available, which means hours are no longer the binding constraint. What is scarce is the judgment that decides whether a competent output is the right one. So sorting by frequency now optimises for a resource you are no longer short of.
Worse, frequency and risk are close to independent. Plenty of repetitive work carries a real call every single time. Plenty of rare work carries none.
The sort: what does a wrong output cost
Take every candidate and ask one question. When this produces a wrong output, and it will, what does the wrongness cost?
| Costs money | Costs trust | |
|---|---|---|
| How you find out | Quickly. A number moves | You mostly do not. Nobody complains |
| Can you undo it | Usually. Refund, redo, resend | No. It has already been experienced |
| Where it shows up | In a report, attached to its cause | Months later, attached to something else |
| What to do | Automate now | Capture the judgment, then automate |
This is a better sort than repetitive-versus-not because it tracks the actual failure mode of automated work. Automated systems do not fail by being slow or inconsistent. They fail by being confidently wrong in a small percentage of cases, and the only thing that matters is what those cases cost.
The two lists in practice
Roughly, and every business differs.
Automate now. Formatting and reformatting. First drafts of things that will be reviewed anyway. Research collation. Transcription and summarising. Data movement between systems. Scheduling. Reporting. Anything where the output has a reviewer downstream, because that reviewer is your correction loop and it already exists.
Capture first. Anything where a customer receives the output directly with no one in between. Anything involving an exception to a normal policy. Pricing that is not from a table. Anything that decides whether a relationship continues. Hiring judgments. Anything where the right answer depends on history the system has no access to.
The dividing line is not difficulty. Several things in the second list are easy. They are in the second list because a wrong one does not come back.
The rule that outranks the sort
One rule sits above all of this and it is worth more than the rest of the page.
Never automate the function you compete on until what makes it win has been captured.
Every business has one or two areas where it is genuinely better than its alternatives, and being better there is why customers choose it. In those areas the quality is not in the process. It is in a set of calls that specific people make, usually without being able to explain them.
Automate that area before capturing those calls and you get exactly what you asked for: the same function, performed to the industry average, faster and cheaper. The cost saving is real and it appears immediately. The thing you lost was the reason anyone chose you, and it does not appear anywhere for a year.
The general form of the rule: build where you win, rent the rest. Rent the average for everything average is fine for, which is most things. Capture and build in the one or two places where average is the thing you were selling against.
Deciding which areas those are is itself a judgment call, and it is a good first entry in whatever you start capturing.
Frequently asked
What should a founder automate first?
Not the repetitive work. Sort every candidate by what a wrong output costs. Where a mistake costs money, it is visible and recoverable, so automate it now. Where a mistake costs trust, it is invisible and not recoverable, so capture the judgment first and automate second. And never automate the function you compete on until what makes it win has been captured.
Why not start with the repetitive work?
Because repetition is a measure of frequency, not of safety, and it was the right rule when the scarce thing was labour hours. Plenty of repetitive work carries a judgment call every time, and plenty of rare work carries none. Frequency tells you where the hours are. It does not tell you where the risk is.
What is the difference between money errors and trust errors?
A money error announces itself. Something is wrong, a number moves, somebody notices, you fix it and you are back. A trust error is silent and cumulative. The customer does not complain, they quietly stop treating you as the one who understands their situation, and by the time it appears in any number the cause is months behind you.
What does 'capture first' actually mean here?
Before the work moves, record what the good calls in that area turn on: the case, the call, the reason, and when the reason stops applying. It is not a documentation project. It is the corrections that are already being made, written down instead of discarded.
What is the one thing never to automate first?
The function you actually compete on, before the judgment inside it has been captured. That function is where your best calls live, and it is the one place where automating before capture converts your advantage into somebody else's baseline.
Where this sits
What happens when the order gets reversed at scale is at Klarna was a sequence failure. What capture actually involves is at how to train AI on your own judgment. The category is defined at what Imprinted AI is.
Related
- Klarna was a sequence failure. What automating the function you compete on, before capture, actually costs.
- Can AI replace a founder's judgment? What accumulates when the order is wrong.
- How do you train AI on your own judgment? What "capture first" means in practice.
- An Imprint vs an SOP. Why the procedure you already wrote does not carry the judgment.
Keep going
The rule at the bottom of this page is the one with the most money attached to it, and the reason it is not obvious is that for twenty years the advice ran the other way.
The full case is The A.I. Business Manifesto. About 23,000 words, free to read on the page, no gate in front of it. If you would rather have the short version, the same page will send you the three fixes and the PDF.
Read The A.I. Business Manifesto
Free either way. Reading it costs nothing and asks nothing.
Last updated: 28 July 2026