The six
Six things decide what you can attempt and finish.
Each has a plain definition, a strong case and a weak case. Each carries a stated weight in your reading, and a first move you can start this week.
Six areas · one bench
How to read this page
Every claim here is settled by a record, not by an adjective.
AGI READY™ reads what your organization can now attempt, carry and finish. Six areas decide that, and each is set out the same way below. What it is, what strong and weak look like, and the record that settles it. Then what it is worth, and the first move to make.
One rule sits over all six. Nothing scores until it has happened once. A plan scores nothing, and neither does a license. What to do this week: mark the two areas below where the record is not on the table yet. Those two are where the next quarter of capability comes from. It takes ten minutes and needs nobody else.
01Ambition · 20 percent
Ambition is the size of what you now set out to do.
Ambition is what you set out to do now, against what you would have set out to do before.
When it is strong
- A leader can name three things attempted this quarter that would have been declined two years ago.
- Each has the name of the person who carried it.
- The declined list is priced again every quarter, at today’s cost of trying.
When it is weak
- There is no declined list, so nobody can say whether the ambition has moved.
- The stated benefit of AI is speed: the plan you already had, faster.
The record that settles itThe attempt list. One page a quarter. Above the line, every project you declined, with its old price. Below the line, what you attempted, and who carried it.
How it is scoredTwenty points, the largest share alongside Direction. What moves it is the attempt rate: declined projects tried this quarter.
The first move. Ask every line leader for ten projects their area declined because trying cost too much. Who: you ask, and every line leader answers. How long: two weeks. You end up with: a dated list of declined projects, each carrying its old price. Done when: every line has answered and every entry names a person.
The attempt list · what came off it
02Individual reach · 15 percent
Reach is how far one person can carry the work alone.
One person · the whole piece of work
Reach is how far one person can carry real work alone, from the idea to the finished thing.
When it is strong
- Work ships that used to need a chain of handoffs, carried end to end by one person.
- People cross the old lines routinely, and nobody thinks it worth mentioning.
- They have a long, logged working relationship with a frontier model, not a license.
When it is weak
- The tools sit inside the old division of labor, so each person does the old job faster.
- The organization chart still describes how the work flows. After two years it should not.
The record that settles itThe reach record. The work that shipped this quarter. Against each item: how many people carried it, how many it needed before, and how long each took.
How it is scoredFifteen points, and only for work that shipped. The comparison is against your own past.
The first move. Pick one thing that shipped this quarter. Write how many people touched it, and how many would have two years ago. Who: the leader whose team shipped it. How long: one hour. You end up with: the first row of the reach record. Done when: the two counts sit side by side, with a date on them.
03Working knowledge · 15 percent
Working knowledge is your own work, written down.
Working knowledge is what you know about your own work, written down well enough to hand part of it over.
When it is strong
- Ask anyone what their team does and you get verbs, not job titles.
- A register carries one row per job whose output drives a decision, with an owner and a review date.
- Beside it sits a private set of real cases with known good answers, in files you own.
When it is weak
- The company describes itself in job titles, and cannot list what it does inside an hour.
- Without cases of your own, nobody can tell whether a model is getting better or worse at your work.
The record that settles itThe register, and the transcripts filed against it. On the table within the hour, every row naming a person and a review date.
How it is scoredFifteen points, and only from transcripts. What moves it is coverage: rows in the register with graded cases filed against them.
The first move. Take ten real tasks from last quarter and hand them to the best model you can get. Grade what comes back against the standard you already use. Who: the person who owns that work, plus a checker who did not choose it. How long: one week. You end up with: ten graded transcripts, filed against the register. Done when: you can name three tasks the machine does to your standard.
The register · one row per job
04Machine capacity · 15 percent
Machine capacity is what actually does the work.
What runs · and what can replace it
Machine capacity is the systems and machine workers that do the work, and how freely they can be changed.
When it is strong
- The company runs machine workers, not only assistants, and jobs run in parallel.
- How much room each gets is written down. It widens on evidence, and narrows after an incident.
- A model’s name lives in one settings file, and a second supplier’s model passed the same tests.
When it is weak
- Model names are written into the software itself, so how long a change would take is a guess.
- How much room a machine gets sits at the supplier’s default, as though that were a fact.
Being able to change supplier protects the ambition. It is not an admission of weakness. One provider’s published retirement page promises at least sixty days’ notice. The same page lists nineteen model versions retired between November 2024 and August 2026.
The record that settles itThe list of what runs, and the timed change. Every model in production, with its supplier, its version and its retirement notice. One named alternative, scored. One rehearsed change, timed.
How it is scoredFifteen points. A change of model scores once it is performed and timed. An estimate scores nothing.
The first move. Ask whoever runs your systems to find model names outside the settings file. Who: whoever runs your systems. How long: one afternoon. You end up with: a dated list of every file to be changed before you can switch supplier. Done when: the list fits on one page and each line carries an owner.
05Direction · 20 percent
Direction is who chooses the work, and who answers for it.
Direction is the judgment that chooses what gets built, and the people who answer for what it does.
Start with what it buys. Because anyone can stop a job, you can let ten run at once without a review board.
A film director never operates the camera and never speaks a line. What the director holds is the shape of the finished thing.
When it is strong
- Every decision that matters has one named person who allowed it and can be asked why.
- Anyone may halt a machine-run job without asking permission, and every halt is logged by name.
- Whoever chose the work is not the person who checks it, and the checker is not the signer.
When it is weak
- Review is a forum that meets monthly, so nobody can stop anything today.
- Approval queues run at a hundred percent, which is attendance rather than judgment.
- Risk is accepted collectively, which is to say nowhere, because no individual signed.
None of this is a brake on ambition. It is what lets a board, a customer and a regulator allow you to run many machines at once. The law asks for the same thing. Whoever supervises a system must have the competence, the training, the authority and the support to disagree with it and stop it. That is Article 26(2) of the EU AI Act, the European Union’s law on AI. It is the duty that falls on the organization using the system, not on the one that built it.
The record that settles itThe signature file. Per job that matters, four things. The named signer. The evidence that they can stop it. The log of the last halt. And what you will not do at any price.
How it is scoredTwenty points, the same share as Ambition. Direction is the only area that counts twice. In the worked example, twelve points of Direction move the whole reading from 68 to 72. No other area moves the total that far for twelve points. Lifting Direction also releases what the other areas are allowed to print.
The first move. Open the register and write one name on every row: the person who can stop that job today. Who: you, with each row’s owner. How long: one afternoon. You end up with: a register where every row carries a name and a way to reach them. Done when: one halt has been logged and you can show the entry.
The shape of the finished thing
06Compounding · 15 percent
Compounding is whether this quarter makes the next one cheaper.
Each turn cheaper than the last
Compounding is how fast you learn from what you just did, and how much of it stays inside your own walls.
When it is strong
- Forward calls are filed with a probability, a date and what would prove them wrong, then scored.
- Dissent is printed by name, and nothing is retouched afterwards.
- Every attempt makes the next cheaper, because the transcripts, the instructions and the graded cases are yours.
When it is weak
- The memory of the place is a set of slide decks.
- The valuable material lives inside a supplier’s product, and leaves when the contract does.
The record that settles itThe record, with the graded cases beside it. Six columns: the call, how likely, by when, what would prove it wrong, who made it, and the score.
How it is scoredFifteen points. Scored calls decide how much room a machine gets. They never decide anyone’s pay or grade.
The first move. Take the three calls you already make every month: the forecast, the hire and the launch. Write each down with a date. Who: whoever already makes each call. How long: thirty minutes. You end up with: the first three rows of the record, each with a date to score it. Done when: the first call has been scored, right or wrong.
07The six together
Here are the six together, with what each one is worth.
Ambition and Direction carry a fifth each, and the four between them fifteen points apiece. The two ends are the ones you have to build yourself.
| Area | Weight | What its share buys | The record that settles it |
|---|---|---|---|
| 01 Ambition | 20% | What you set out to do | The attempt list, priced again each quarter |
| 02 Individual reach | 15% | How far one person carries it | The reach record, with the names on it |
| 03 Working knowledge | 15% | What can be handed over | The register, and the transcripts behind it |
| 04 Machine capacity | 15% | What does the work | The list of what runs, and the timed change |
| 05 Direction | 20% | Who chooses, and who answers | The signature file, and one logged halt |
| 06 Compounding | 15% | Whether it gets cheaper each time | The record, and the private set of cases |
Areas one and two set the size of the ambition. Three and four are the machinery. Five is who answers, and six is what makes the next attempt cheaper.
The weights are ours, and no standards body has certified them. They are printed here so that anyone who disagrees can say so precisely.
What to do this week. Print the table, and beside each row write the record you could produce today. Twenty minutes, alone or with your leadership team. The blank rows are your first plan, and the route is where they get filled in.