The Unbiased Master Plan

The mission

Our mission is to put a genius or two to work beside every human on Earth, regardless of what they earn. How? We'll start by making AI providers compete for every request. Why? Because human potential is spread evenly around the world, and opportunity isn't.

We're publishing the whole plan because it needs time, and it works better the more people understand it. Somebody is sitting in the shade today only because somebody else planted a tree many years ago.

The last founder who published one of these went on to build several trillion-dollar companies.

Saying the quiet part out loud

A new frontier model ships every few weeks. Each one leapfrogs the last on some benchmark, trails it on another, and prices its tokens differently. If you run AI in production, you now employ smart people whose actual job is keeping up - rereading model cards, rerunning evals, migrating prompts, renegotiating commitments. Indecision has become one of the largest line items in your AI budget, and it appears on no vendor's invoice.

Here's the uncomfortable part: everyone at the table profits from your confusion. Don't take our word for it - ask them.

• Ask your lab why it never recommends a rival's model, even when the rival is better.
• Ask your cloud why the savings only appear alongside a big commitment.
• Ask your consultants what happens when your model decision becomes simple.

They aren't lying, they're just conflicted. Not one of them gets paid to make intelligence cheap. But we do.

What we're building

The Pareto frontier is the curve formed by plotting options against their performance trade-offs. One example is quality versus cost: the curve represents the best possible answer available at any given price point.

We named our model Pareto because it embodies this concept: it is the line itself. The goal is for every request to go to whichever set of blended models does the job best for the least money. Our target is frontier-level performance at about a quarter of the price; the public model card and reproducible benchmarks will be the proof.

And the line moves so that a fixed capability gets meaningfully cheaper every year. Pareto is being built to ride (and accelerate) the whole descent - no target price, just lower. We intend to do the keeping-up, automatically, per request, forever; you keep the difference.

That should sound unbelievable. Good. The public Pareto model card will hold the full specs and published benchmarks - and we will invite others to run them.

How you will check our work

The Unbiased API is being built so you can point it at your traffic and see which model won, and what it should have cost. When comparison mode is available, you will be able to run Pareto head-to-head against third-party inference providers of your choice and generate a report showing you:

• How often Pareto would have given the same answer as the competing model.
• When the two models disagreed, how often the competing model preferred Pareto's response upon reviewing it.
• How much Pareto's completion cost in comparison to the competing model.

Over time, the goal is to produce a clear picture of the optimal model for each of your inference workloads, on a cost-adjusted basis.

Why you can trust a middleman - structurally, not personally

The honest objection to everything above: why would the company in the middle pass the savings along instead of keeping them?

Costco answered that question so well their name became shorthand for trust: a membership fee at the door, a hard cap on the margin of everything in the warehouse, and a business that makes its money from long-term retention instead of the markup. Vanguard ran the same play on Wall Street; Costco for investing. One structural promise, two forms: we win only when you pay less.

We're building the Costco of AI. Members pay a fee at the door.

Inside, the plan is for the aisles to be honest by construction. Every vendor's model will pass through untouched - your keys, their prices, zero markup. And like Costco, we will stock a house brand. Kirkland exists for one reason: when the store thinks it can make something better for less, it takes responsibility for production and quality. Pareto is ours, and the shelf will keep us honest; the name brands will be right there.

Membership will let you walk the aisles: every price visible, every model comparable, the samples free to try. The comparisons will be the personal advisor walking beside you - pointing at your cart and telling you, item by item, what you should have paid. The guarantee we intend to make is simple: your membership pays for itself or it's free. Every month, your statement will show what you saved; any month the membership did not cover its own cost, that month's fee will come back as credit.

We'd rather save you dollars than pennies; the statement is the proof.

Ten years from now

For centuries, opportunity has always been a lottery nobody chose to enter, and the ticket is where you were born. We're making sure there isn't a lottery in 2036.

Today, most humans still haven't tried AI. Those just now meeting it are likely on a phone, in a place where a frontier subscription costs a month's wages. We find that problem both unacceptable and solvable. A price that falls tenfold a year is compound interest running in reverse: patient, relentless, and working for whoever waits.

Our job is to make sure nobody waits longer than they have to. By 2036, we'd like everyone to be working with a genius or two, regardless of what they earn - the kind that once billed by the hour to the world's richest firms, seated instead at every kitchen table on Earth.

Picture a mechanic in Nairobi with the same intelligence a hedge fund rents by the hour. She doesn't need smarter answers - she needs the bills collected, parts sourced, and customers answered while she's under the car. A colleague, not an oracle. Delete that second shift she doesn't need to work. That's the difference between access and income, and income was always the point.