The equation has been rewritten twice since May. Both earlier versions are on this page, because a studio that quietly swaps its own thesis and pretends the old one never existed has learned nothing it can reuse.
Maintained as a versioned document. Every revision is dated, and no version is removed from the record when it is replaced.
Each revision moved the centre of gravity one step closer to where the variance actually sits. The first two could only be agreed with. The third can be diagnosed, located and climbed.
AI is the base. You are the exponent. What you bring to the equation is what compounds.
What it got rightThe human is not optional. At a moment when most of the market was arguing about AI replacing people, the mark insisted a person was in the equation at all. That instinct was correct, and the studio was built on it.
What forced the changeIt put the machine at the centre. Read literally, AI is the subject and the human is the modifier: the value sits in the AI and you improve it. That is the opposite of the position we actually hold.
You are the base. AI is the exponent. The flip.
What it got rightIt put the human at the centre and made AI the amplifier. That is the correct direction of leverage, and it is the half of the correction that survived into 3.0.
What forced the changeIt fixed the relationship. It asserted that the human raises the machine, always, for everyone. If the exponent were automatic there would be nothing to teach. It also collapsed two different AIs into one term and hid the fact that they do different jobs and fail in different ways.
Written and superseded inside the same working session. It never reached the site, and the speed of its replacement is the clearest evidence that the second correction was the real one.
Observation, not theory. How people apply AI to expertise is hyper-variable. Two leaders with the same domain knowledge and the same subscription produce output that differs by an order of magnitude. Neither 1.0 nor 2.0 could account for that, because both fixed the relationship in advance.
What it addsTwo distinct AIs rather than one. An operator that is earned rather than assumed. A strength setting most people never touch. And an output term, which forces the honest question about what the whole arrangement is actually producing.
What it costsSimplicity. The first two fitted on a business card. This one asks for a few seconds before it pays. It is the right trade, because the earlier versions could only be agreed with and this one can be used.
One is pointed at the subject, to sharpen what you know. The other is pointed at the work, and its strength is a setting you choose. Between them sits the operator, and that is where two people with identical expertise and identical subscriptions separate completely.
What you actually know about the subject in front of you. Judgement, know-how, the pattern library you built the hard way. Nothing downstream manufactures this.
Language models and code-based application pointed at the subject to sharpen the knowledge itself. The output of this pass is not a document. It is a sharper you.
How the sharpened bracket is applied to the machine. Not selected, earned by practice. This is the variable, and it is the only term that is genuinely yours.
How much machine you actually bring. Model selection, token spend, iterations, engine fit. A dial, not a constant, and most people leave it near the bottom without knowing it is there.
AI sits beside the work. Tasks in, answers out, contributors stay separate. Output is the sum of the parts and never more. Nothing accrues. Stop on Friday and Monday looks exactly as it did before.
AI acts on the whole bracket. Context, then more context, then a role, then let it interview you before it produces anything. Your expertise stops being an input and becomes a coefficient.
AI applied to the system that makes the work, not the task and not the method. Each pass raises the base for the next one. The only tier where output outlives the session that produced it.
Reaching the top operator is not automatically a good outcome, and the arithmetic is honest about it. Exponentiation amplifies whatever is inside the bracket, in whichever direction the bracket already points. A strong bracket runs away upward. A thin one collapses, faster than it would have decayed on its own.
Exponential application of shallow expertise does not produce shallow output. It produces confident, fluent, well-formatted output that is wrong, at volume.
Which is why the first AI has to do its work before the second one is turned up. Sharpen what you know, then reach for the operator. In that order.
| Version | Mark | Dates | Centre | Status |
|---|---|---|---|---|
| 1.0 | AIHI | 27 May 2026 to Aug 2026 | The machine | Published, superseded |
| 2.0 | HIAI | 5 Aug 2026 | The human | Never published, superseded same day |
| 3.0 | (HI + AI)+×^AI | 5 Aug 2026 to current | The interaction | Current |
No version is removed from this record when it is replaced. Superseded is a status, not a reason to delete the evidence.