We are measuring the wrong thing

Originally published on billfaruki.substack.com on June 5, 2026. Republished on MindHYVE.ai with light canon edits.

Almost every conversation about AI right now is about capability. How long a model can run on its own before it needs a human. How much harder a problem it can solve this year than last. Whether the curve is still bending up or starting to flatten.

It's a real conversation. The progress is real, and the people tracking it closely are not wrong to be impressed. It's just the wrong thing to lead with — because it answers a question that was never the one that decides what this technology does to the world.

The question that decides that is not how capable AI becomes.

It's who the capability reaches.

Begin with the most conservative case anyone can make.

One leading model lab — a company whose entire business is the next capability jump — recently made the point plainly: even if every model were frozen exactly as it is today, the world would still change enormously. Not because of what comes next, but because of what already exists, still working its way into the economy. A company of a hundred people starts doing the work of a thousand because each person now sits on top of a stack of agents doing the work beneath them.

Sit with that for a second, because it's a stronger claim than it first looks.

It means you don't have to win a bet on the next breakthrough for the stakes to be enormous. The capability already in the world, fully distributed, is enough to reorganize how work happens. Freeze everything today and the transformation still comes.

So the real question was never only how much more capable these systems get.

It's where the capability we already have actually lands.

The honest answer is: not evenly.

The reasoning the field keeps making more powerful has mostly reached the people who were already well served. The large firm that can staff an AI team. The institution with a procurement budget. The professional who can pay for the frontier tier.

That's not a moral failing of the labs building it. It's just how diffusion works when no one is steering it. Capability flows along the paths that already exist, and those paths run toward the people who already had the most.

Which leaves the other side of the line.

The doctor working a clinic with no specialist down the hall to consult. The lawyer with no firm behind her and no senior partner to check the brief. The teacher with forty kids and no aide.

These are not edge cases. Numerically, they are most of the profession. Most doctors, most lawyers, most teachers, in most of the world, practice without the institutional weight that the elite version of their job takes for granted.

The reasoning that used to be locked inside those elite institutions is exactly the thing that could now reach them.

That's the actual frontier.

It is not getting the same attention as the capability frontier, and that gap is the whole problem.

Here is the part the field should sit with.

We have built an elaborate, shared, obsessively tracked vocabulary for measuring how smart these systems are. Benchmark scores. Task-completion length. Eval suites that get harder every quarter. We argue about the numbers, we publish them, and we orient entire roadmaps around moving them.

We have almost nothing equivalent for reach.

No shared metric that asks the only question that matters on the other side: did the capability get to the practitioner who didn't have access to it before?

Is the doctor in the clinic with no specialist actually reasoning at a higher level today than last year — or did all the gains accrue, again, to the people who were already ahead?

This matters because of a rule that holds everywhere:

A field that measures only intelligence will keep producing intelligence and, by default, keep delivering it to the already served. Not out of malice. Out of inattention.

You don't improve what you don't count.

Right now, we are not counting reach.

Taking reach seriously does not mean counting users.

User totals are the vanity version of this — easy to inflate, easy to confuse activity with impact, and they tell you nothing about whether the capability reached anyone who lacked it before.

A million sign-ups concentrated among people who already had three other tools is not reach.

It's redistribution among the comfortable.

The real measure is harder and more specific.

It asks: which practitioner, in which setting, can now do what they previously could not — and did they lack it before?

The honest, defensible claim is not that any practitioner with this technology becomes the equal of the most elite version of their profession. That's the version that sounds good in a keynote and falls apart on a single bad outcome.

The defensible claim is narrower — and, I would argue, more important:

Not the credential.

Not the brand.

The reasoning.

That's the thing that was locked up, and that's the thing that can now travel.

When that's the goal, the work looks different.

You don't build for the user who already has everything. You build for the one the existing paths never reach — and you have to go to where they are, because diffusion alone won't carry it there.

Most people think the story of this moment is how fast it's moving.

I think the story is who it reaches while it moves.

Speed is the headline.

Reach is the substance.

This is the problem I've spent the last few years building a company around, so I won't pretend to be neutral about it.

We operate across three continents — Newport Beach, Islamabad, and Nairobi — for a simple reason: if this only works in San Francisco, it was never equalization to begin with. It was just the same advantage, delivered faster, to the same people.

The capability frontier will take care of itself. There are brilliant people and enormous resources pushing it forward every day, and they don't need my help.

The reach frontier doesn't have that.

It's underfunded, under-measured, and under-discussed relative to what it decides.

I think that's the wrong allocation of attention.

I think correcting it is one of the most important things happening in this field — even though almost no one is keeping score.

So start keeping score.

First published on Bill Faruki's Substack on June 5, 2026. Lightly edited for house style. Read the original.