Home ENDE

The method

Which part of this week is machinery pretending to be judgement?

Every person I have worked with arrived with a different job title and the same sentence: one part of my week has become unbearable. The profession changes. The shape of the problem almost never does.

Everything I build follows from the question above. I do not ask what someone wants automated, because almost nobody knows — they know what hurts, not where the seam is. Finding the seam is the work. Building on the right side of it is the easy part.

How I tell one from the other.

Machinery — all four are true

  1. It repeats. More than once a week, in roughly the same shape.
  2. It is rule-shaped. If you had to explain how you do it, you could — in sentences, not in gestures.
  3. It is verifiable. A wrong output is visibly wrong to you in seconds.
  4. It is low-consequence when caught. If the machine gets it wrong and you catch it, nothing is lost but a minute.

Judgement — any one is true

  1. You are liable for it. Your name, licence or signature is on it.
  2. Taste is the product. People pay you precisely because the view is yours.
  3. The relationship is the product. A patient, a student, a client, a reader.
  4. It is a real decision under uncertainty. Not a lookup with extra steps.

I build up to the seam. I never build across it. Nearly every failed AI project I have been asked to rescue failed because somebody built across the seam, and the person quietly stopped trusting the output.

How it works

Four phases, every time.

  1. I sit in the week.

    Two to five days watching the actual working week. Not a workshop, not a questionnaire — the mailbox, the folder names, the sticky notes, the workaround you are embarrassed about. The workaround is usually where the system goes.

  2. I name the seam out loud.

    One page saying what is machinery and what is judgement, read back to you. If you flinch at something I called machinery, you are right and I am wrong. Your flinch is data. That page is the contract.

  3. I build the smallest thing that removes the volume.

    Not a platform. The one system that takes the repetitive mass off you. It ships in weeks, not quarters, and it does one thing so completely that you stop thinking about it.

  4. I hand over the controls and leave.

    You must be able to change the rules without me: the filters, the thresholds, the templates, the reject reasons. If a system requires me to stay, I built it wrong.

When I say no.

I say no out loud, and I keep the relationship. Saying no is most of the reason people send me their friends.

Volume is too low.
Under roughly an hour a week, the maintenance costs more than the task.
Variance is too high.
If every instance is genuinely different, there is no machinery to lift — there is only skill, and skill is not a bottleneck to remove.
Liability is too concentrated.
If one wrong output ends a career, the machine gets to prepare and nothing else.
A boring tool already exists.
A calendar, a form, a shared spreadsheet, a template. I will set that up and bill for the afternoon, not the quarter.
It is a decision, not a workload.
Some people do not need a system. They need to decide something they have been avoiding. No software fixes that.

Two registers.

I work inside a company, where I own and run a production system used by a real legal department, and I work one-to-one with a yoga teacher or a chef at their kitchen table. The company work teaches me what survives contact with audits, handovers, and people who did not ask for it. The one-to-one work keeps me honest about how little most people actually need.

Get in touch

Tell me what your week looks like.

I'll tell you honestly which parts you can hand over — and which you can't.

Message me — the first conversation is free See it applied, field by field