Skip to content
howreadyfor.aihowreadyfor.ai

The AI Readiness self-assessment

You think you are further along on AI than you are.

Most people do. And AI can generate code faster than your system can safely absorb it.

Across five major AI coding assistants, 17% to 29% of commits introduced at least one detectable issue. Almost 23% of the issues tracked remained in the codebase.Liu et al. · 2026 302,579 AI-authored commits across 6,299 repositories

Twenty questions about the system you really run: the repositories, the pipeline, the tests, the branches. One at a time, about half an hour, and it will not flatter you.

01

You are already making readiness decisions.

Approving a tool, widening a pilot, telling a board that agents will lift delivery next year: each of those decisions assumes your system can absorb the change. None of them measures whether it can.

Unmeasured is not the same as wrong. It means there is nothing to argue with, and when nobody can argue, the most confident opinion in the room wins by default. The error also has a direction. You know your best-run services best, because those are the ones that get demonstrated. In engagements, the thing that caps a system is almost never the thing I am shown first. An agent has no idea which part of your estate you are proud of, so it reaches the parts you are not proud of just as fast.

Half an hour buys you a number, the one dimension holding that number down, and a ladder published in advance, so anyone who disagrees has to argue with the ladder instead of with you. If it is better than you feared, you spent half an hour. If it is worse, you found out before you scaled.

02

The floor is the score.

The assessment breaks your system into 20 separate dimensions and scores each one from 1 to 5. Repository structure. CI reliability. Test meaningfulness. How old your oldest branch is. Seventeen more.

Your floor is the lowest of those 20 scores, and your floor is your readiness. Not the average, not a weighted composite, not a percentage of a maximum.

You already rate systems this way. A chain, a pipeline, a load path: the weakest part sets the rating and nobody averages the rest. Readiness is the same shape.

Which is why most floors are low, and that is arithmetic rather than an accusation. One weak dimension out of 20 caps the result, however good the other nineteen are. A minimum over twenty dimensions behaves the way a minimum behaves.

One dimension at 1. The average still reads 2.7.

Floor · 1The verdict. Requirements quality is holding it there.Scoped to: one product (Payments)
An average · 2.7What a mean would report instead.
12345
A worked exampleTwenty dimensions, grouped into five clusters, scored for one declared scope. Nobody scored this; these twenty numbers are illustrative. What is not illustrative is the shape. The solid line sits at or below every single column, because that is what a minimum does. An average cannot make that promise.The twenty scores in this worked example are: Repository structure, 3; Cognitive load per module, 3; Domain language consistency, 2; Semantic drift, 2; CI reliability, 4; Test meaningfulness, 2; Change absorption, 3; Technical debt, 2; Trunk-based development, 4; Velocity and the constraint, 3; Organizational design, 3; Code ownership and review routing, 3; Decision memory (ADRs), 2; Requirements quality, 1; Security assumptions, 3; Documentation integrity, 2; Hard guardrails vs cultural agreements, 3; Architecture enforcement in CI, 4; Dependency hygiene, 2; Release discipline, 3.
03

Five levels, and one question decides which.

The same ladder applies to all 20 dimensions. Nothing derived it. It is what climbing looks like from inside real organisations.

  1. 1

    Blind

    The capability does not exist. No standard, no metric, no named structure. You are blind here.

  2. 2

    Tacit

    It happens, but inconsistently, by individual effort. It lives in someone's head; the system does not do it.

  3. 3

    Visible but inert

    Written down, computable, or diagrammed, but nothing acts on it automatically. The dangerous middle: it feels safe, and nothing in it stops an agent.

  4. 4

    Acted on

    The system itself acts, no human choice required. A machine blocks it, a metric gates a decision. It cannot quietly slip.

  5. 5

    Durable

    Acted on and self-maintaining. It survives acceleration, staff turnover, and next quarter's model upgrade.

Past this line the system acts. Before it, someone has to remember.

That line is the whole ladder. One question decides which side of it you are on: does the system itself act on this, or does a human have to choose to? Levels 1 to 3 need a person to choose. Levels 4 and 5 do not.

04

The kind of question it asks.

Four of the 20, so you know what you would be answering.

  1. 01How old is your oldest active branch?
  2. 02If your pipeline went red right now, would your team investigate, or just rerun it?
  3. 03Of the dependencies you run in production, how many did a human actually choose?
  4. 04If AI doubles each team's throughput, do the queues shrink or grow?

There is no clever scoring behind them. You read five descriptions of a real situation and pick the one you recognise. Each of the five carries a tell: the thing that would settle it, so the level you land on is something you checked rather than something you agreed with. That is slower than clicking a radio button, and it is the only reason the number means anything.

All 20, in 5 clusters, with the article behind each one, are on the method page.

05

The rules it is built under.

Half an hour of honest answers, and nothing else asked of you. Three of these are checked on every build, and a build that breaks one does not ship. The other two are true because of how the site is built. That is a weaker guarantee, and the wording below says so.

No email, no account, no newsletter
Checked on every build. It searches the rendered pages for an email input, or for a field named email, and fails if it finds either. There is nowhere to put an address, because the markup cannot contain one.
The PDF will never be locked behind a form
True by construction rather than by a check. The report is rendered from your own answers, in your own browser. There is no server withholding it and no step between finishing and exporting, so there is nothing to put a form in front of.
No price, and nothing to buy
Checked on every build. A euro amount written out anywhere in the rendered site fails it. Nothing is sold here, so a number with a currency sign in front of it is a defect rather than a decision.
Nothing leaves your browser until you finish
True by construction rather than by a check. Your answers live in your own browser and never expire. There is no session to expire 24 minutes in, because there is no session. What is sent when you complete an assessment is itemised in the privacy notice.
No spreadsheet to download
Checked on every build. A link to a spreadsheet file fails it. The spreadsheet this replaces is already public elsewhere, and offering it here would concede that this instrument is no better than a spreadsheet.

A rule nothing enforces is a level 3, and a level 3 is the dangerous middle.

Score it

Somebody is going to ask you what agents do for delivery next year.

Answer it with a number instead of a position. Twenty questions about the system you really run, one at a time, about half an hour. You leave with your floor, the one dimension holding it there, and where to start.

No account, no email, nothing sent anywhere: your answers stay in your own browser. And the ladder is published here in advance, so when you put the number in front of somebody who would rather it were higher, the argument is with the ladder rather than with you.