The AI Armor · t = now − 2020

The tool stops needing the hand.

A hammer cannot drive a nail by itself. Its output is bounded by the arm that swings it, so the person stays the deciding term. The AI armor is not a hammer. It is a layer that multiplies what a person can do — and the multiplier grows on its own.

That single difference changes the arithmetic. Once the armor can prompt itself, it no longer depends on how skilled its user is. Knowledge of the tool stops being an advantage: the user only has to be able to state what they want. The armor supplies the rest. And because it supplies the rest, the share of the result that is you keeps shrinking — even while the total keeps rising.

This is what suppression looks like: not a loss of ability, but a loss of proportion. The person keeps improving. The armor improves far faster. On 1 January 2027 the two cross, and after that the human contribution is a rounding error inside a result the armor is mostly responsible for.

Why this page exists

To make a specific claim inspectable: an AI layer that multiplies human capability will, by the shape of that multiplication alone, reduce the human share of the outcome toward zero. Nothing here requires the AI to be hostile or the human to get worse. The suppression is structural. The chart below is the argument; the slider is there so you can test how much the result depends on the one free parameter.

humany = b · (1 + 0.1t) armorx = 1 + cap((t/7)2 + 0.5t, V) productz = x · y
computing… A = B = 1 · crossover pinned to 1 Jan 2027 · growth-rate cap V = 100

The argument

four steps, no assumptions about intent

01

A hammer is bounded by the arm

With a hammer, output is roughly the force of the swing. The tool adds almost nothing of its own, so the skilled person and the unskilled person get very different results. Skill is the dominant term, and it stays dominant no matter how long you use the hammer. This is why "human capability × tool capability" behaves sensibly for ordinary tools: the tool's factor is close to constant.

02

The armor can prompt itself

An AI layer is different in one decisive way: it generates the next step without waiting for the user's skill to generate it. The user no longer needs to know how to do the thing — only how to describe it. The moment that is true, the tool's contribution stops being a constant and starts compounding on its own. The equation becomes result = armor × human, where the first factor is now the one that grows.

03

Knowledge stops being leverage

When the armor can supply expertise, knowing more about the domain buys you less and less. A person who has spent years learning a field and a person who can articulate what they want end up with similar output, because the armor fills the gap for both. This is the uncomfortable part of the idea: the thing we usually call "being good at this" is exactly what the armor absorbs first. The advantage that survives is the ability to state intent clearly, not the ability to execute.

04

So the human share falls, not the human

Note what is not happening. Human capability is not decreasing — in this model it rises continuously. The product is not shrinking — it rises faster than either factor alone. What collapses is the ratio. The human share of the total goes from nearly all of it to a fraction of a percent, while both the person and the total improve. That is what makes the suppression easy to miss: every individual number looks like progress.

Capability over time

Factors on the left log axis · product on the right log axis · t = years since 1 Jan 2020

log scale · ←→ to inspect
velocity capV = 100
cap binds — armor then grows +V / year minimum product —

Interactive chart. Use left and right arrow keys to move a time cursor; shift plus arrow moves faster; Home and End jump to the range edges; Escape clears the cursor.

What the arithmetic says

computed live from V

Reading the model

three terms, one cap, one product

The equations

A = B = 1. The cap is the only free parameter.

t = now − 2020

Years since 1 Jan 2020, fractional.

y = b · (1 + 0.1t)

Human capability. Grows gently and never falls.

x = 1 + cap(a·(t/7)2+0.5t, V)

Armor multiplier. Equals 1 at t = 0, reaches 2 at t = 7, then grows at slope V once its growth rate would exceed V.

z = x · y

The product. Both factors are ≥ 1, so z ≥ 1 for every t.

sharehuman = y / z = 1 / x

Since z = x·y, the human share is exactly the reciprocal of the armor multiplier.

How to read it

At V = 100.

  • The crossover is pinned. x reaches 2 at t = 7, so the human share crosses 50% on 1 Jan 2027. Before that the person holds the larger portion; after it, the armor takes over.
  • The cap binds later. Whenever the growth rate would exceed V, the armor switches to a straight line of slope V. At V = 100 that happens on 15 May 2032.
  • Human share = 1/x. The human portion halves every time the armor multiplier doubles: 50% at x = 2, 25% at x = 4, 10% at x = 10.
  • z ≥ 1. Both factors are at least 1, so the product never falls below the human baseline.
Data table — exact values
Values at selected times for the current V.
datetarmor xhuman yproduct zhuman sharex / y

What this means

the claim, stated plainly

The interesting thing about the armor is not that it makes a person stronger. It is that it eventually stops needing the person to be strong. A hammer needs an arm; the armor needs an intention. And once a tool needs only an intention, the skill of the person holding it is no longer the thing being multiplied — the armor's own capability is.

That is why the equation doesn't behave like a normal tool. With a hammer, human × tool works because the tool term is roughly fixed: it is a multiplier near one. Here the multiplier rises without bound and without instruction, so the product grows while the human share collapses. The person is not defeated. The person is diluted.

The uncomfortable conclusion is that this does not require anyone to be replaced, deskilled, or made worse off. Every quantity a person cares about — their own capability, the total output, the quality of the result — can be improving at the same time that their share of the outcome approaches zero. Suppression of this kind does not announce itself. It looks like a chart where everyone is winning.

The person stays in the loop. They just stop being the reason the loop produces anything.

Where it surfaces

the dilution, translated into markets

None of this stays abstract for long. Once the armor is what produces the work, it is also what gets sold — and the price of a capability collapses when the thing that supplies it stops being scarce. The armor becomes accessible to almost anyone, and the market does what it always does when supply of a resource explodes.

  1. The armor gets cheap

    What used to require a team, a budget and years of accumulated skill now requires one person, a clear description of the outcome, and a subscription. The cost of producing capability falls toward the cost of stating intent.

  2. Competition arrives from everywhere

    A single person with the armor can ship a whole product. They no longer need to hire the team that used to be mandatory, or license the expertise that used to be proprietary. Every customer becomes a plausible competitor, because the barrier that stopped them was never ambition — it was capability.

  3. Prices fall toward that new cost

    With supply of execution effectively unbounded, prices stop reflecting how hard something was to build and start reflecting how cheaply it can be rebuilt. The margin that came from difficulty evaporates, because difficulty was the product.

  4. Firms shrink without doing anything wrong

    Companies built on selling execution contract — not because their work got worse, but because their buyer became their rival. A firm that employed forty people to build something now competes with customers who can build it themselves. Nobody at that firm was replaced. The market simply stopped needing forty people to produce the same thing.

  5. Knowledge workers lose the premium on knowledge

    Software developers are the clearest case. The profession was valuable because knowing how to build software was scarce and hard-won. When the armor can supply the "how", the scarce input becomes the "what" — the judgement about what should exist and why. Developers do not get replaced by better developers. They get displaced by the people who used to hire them.

This is the same chart, viewed from a different axis. The human share of the work falls, then the human share of the price falls, then the human share of the headcount follows. Each step looks like a local market story — a cheaper tool, a crowded field, a company that lost its edge. Taken together they are one curve.

The developer is not replaced by a better developer. The developer is replaced by the customer.