How Boring Is Your AI?

Hypothesis

When managers rate their AI work against situations they can actually see, their biggest gaps turn up in the boring parts: a mapped workflow, someone who checks the output, a number agreed before the build. Not in the tools.

Most AI readiness checks ask how strongly you agree with statements like “we have a clear AI strategy.” Everyone agrees a little, and nobody learns anything. The premise here is that readiness shows in ordinary work: who checks the output, what happens to a pilot that misses its number, whether anyone could list the AI tools in use.

Method

A free self-assessment that takes about five minutes.

• Six dimensions, three questions each, each built on one of the frameworks: prioritization and risk tiers, workflow readiness, governance and intake, trust and verification, people and judgment, and measurement and value.
• Every answer is a situation, not a rating. Each question has four options, from “people work around it, and nobody owns fixing it” to “we fix the cause first”, scored 0 to 3. The instructions ask for an ordinary Tuesday, not your best day, and the lower answer when you’re torn.
• Four stages, on this site’s metaphor: Magic, Alchemy, Chemistry, Boring. The stage comes from the average answer, with one rule taken from the philosophy: the overall stage can sit at most one step above the weakest dimension, because one weak link is enough to keep AI from being dependable.
• The result names the three biggest gaps, each with the answer given, one concrete next step and a link to the framework behind it.
• It works for a department or a whole company. The choice changes the wording of a few questions, not the structure.

How the tool keeps itself honest:

• A snapshot, not a benchmark. The result says so plainly. There are no percentiles, no industry averages and no comparisons with other companies, because there’s no data behind any of them.
• No sign-up, no stored answers. A result lives in its own link, so sharing it means sharing the link. The optional email form stays off until a real service is behind it, and its box for occasional notes is never ticked for you.
• The scoring is written down. “How the score works” on the results page spells out every rule, and each rule is covered by tests.

Result

Live on this site, under Tools, since October 2026. Answers stay in each person’s browser and in their own link, so there’s no central dataset, by design.

What building it showed:

• Writing the options is the hard part. An agree scale writes itself. Four situations per question, from least to most ready, take a clear picture of what “ready” looks like on an ordinary day.
• The philosophy changed the arithmetic. A plain average lets a strong dimension hide a weak one. Holding the stage to one step above the weakest dimension makes the score say what the philosophy says.
• Honest scope makes the result more useful. “Self-reported snapshot, not a benchmark” turns the result into the start of a conversation instead of a grade.

What I’d change

• Test the hypothesis without breaking privacy. Nothing is stored, so the pattern across teams stays invisible. An opt-in, anonymous tally of dimension scores would show whether the gaps really cluster in the boring parts.
• Check the cut-offs against real teams. The stage thresholds split the scale into equal quarters, a starting assumption. A few teams assessed from outside would show whether “Chemistry” means the same thing to everyone.
• Compare answers within a team. The most useful version may be several people on one team taking it and talking about where their answers differ.