Build this?

Free tool

Should this be built?

Eight questions I ask in the first hour of any AI product conversation — one for each of the eight rules I hold AI products to. Answer them honestly about one idea and you get a verdict, the rules that decided it, and a record you can send to whoever has to agree with you.

It will say no. About half of the ideas I am shown should not be built yet, and the tool is calibrated the same way.

What it is
My scoring rules for “is this worth building”, written down and made usable.
What is real
The questions, the readings and the verdicts. They are the ones I use.
Built on
My eight principles, one question each. Which question tests which rule.
What it is not
An analysis of your situation. It cannot see your organisation; a two-week assessment can.
Where it runs
In your browser. No server, no model, no account. Nothing you enter is sent anywhere.
Built in
One day · 14KB inline, most of it the copy · no dependencies

Should this be built?

Eight questions

01What would this replace?

Tests principle 01 · Do not pave the cow path.

02Who acts on what it produces?

Tests principle 02 · Design the decision, not the dashboard.

03How did this get decided?

Tests principle 04 · AI increases the value of judgment.

04When it is wrong, what happens?

Tests principle 06 · Explain outcomes, not implementation theatre.

05How bad is a wrong answer?

Tests principle 05 · Scale autonomy with the blast radius.

06Does the data it needs exist?

Tests principle 07 · Challenge the agent.

07What will people stop doing because of it?

Tests principle 03 · Adoption is part of the product.

08In ninety days, how would you know it worked?

Tests principle 08 · Build from the outcome backwards.

Why not just ask a chatbot

A model tells you what you asked for. This tells you what I think.

It has a position.

A model gives an agreeable answer shaped by how the question was put. This scores every idea against the same eight rules — the ones on the principles page — in the same order, and prints which rule decided it. Same answers, same verdict, every time.

It asks what you did not.

Who overrules it. What people will stop doing. How you would know in ninety days. Those questions are the assessment; a prompt only asks what you already thought of.

Nothing leaves the page.

You can describe a confidential idea here without it going to anyone’s server, including mine. That matters more than usual for the people I tend to work with.

It is a rubric, not a consultant.

It cannot see your data, your politics or your team. It gives you the frame and a first verdict. The rest is two weeks at the desks of the people doing the work, which is what I do.

Where the questions come from

One question per rule.

The eight questions are the eight principles turned into something you can answer about a real idea. Each verdict names the rule that decided it, so you can argue with the rule rather than the tool.

  1. 01What would this replace?01Do not pave the cow path.
  2. 02Who acts on what it produces?02Design the decision, not the dashboard.
  3. 03How did this get decided?04AI increases the value of judgment.
  4. 04When it is wrong, what happens?06Explain outcomes, not implementation theatre.
  5. 05How bad is a wrong answer?05Scale autonomy with the blast radius.
  6. 06Does the data it needs exist?07Challenge the agent.
  7. 07What will people stop doing because of it?03Adoption is part of the product.
  8. 08In ninety days, how would you know it worked?08Build from the outcome backwards.

Got a verdict you disagree with?

Send me the record