Token capital is the new IP of the firm.

This platform change is unlike the ones before it.

Every previous platform shift — mainframes, PCs, the web, mobile, cloud — did roughly the same thing in different costumes. It gave humans better tools to do work they already understood. The work itself stayed human. The tool got faster.

This shift is different. For the first time, the tool doesn’t just execute work — it learns from it. Every interaction between a person and an AI system produces a residue: a judgment, a correction, a preference, a pattern. That residue is the most valuable thing a firm produces this decade, and almost no one is keeping it.

The interesting question is no longer which model is the best. Models will leapfrog each other on a six-month cycle until the differences stop mattering. The interesting question is whether your firm owns the learning loop that sits on top of those models — the system that turns every interaction into durable institutional knowledge.

Without that loop, the model is a rented brain. With it, the model is replaceable infrastructure, and the brain is yours.

Human capital is what you’ve always had. Token capital is what you build next.

Human capital is what every firm has been accumulating since it was founded: the knowledge, judgment, relationships, ingenuity, and pattern recognition of the people inside it. It walks in on Monday and out on Friday. When someone leaves, some of it leaves with them.

Token capital is new. It’s the AI capability a firm builds and owns — the workflows, domain knowledge, accumulated judgment, and evaluations that have been encoded into a system that improves with every use. Unlike human capital, token capital doesn’t walk out the door. It compounds.

The instinct is to assume one replaces the other. It doesn’t. Human capital becomes more valuable as token capital grows, not less. Token capital is dead matter without human direction — compute running in circles. Humans set the ambitious goals, connect the dots across domains, and recognize the patterns that matter. The loop is what makes both kinds of capital compound together.

Most teams are spending tokens like they spend electricity. They shouldn’t be.

Tokens are a cost, not an asset. A team spends a hundred dollars in API calls today, gets some output, and ships it. Tomorrow the same team spends another hundred dollars on a near-identical question and gets nearly the same output. None of yesterday’s spend made today’s spend cheaper or better. The token bill compounds. The value doesn’t.

Learning lives with the individual, not the team. The person who got really good at prompting the model for variance commentary has all that skill in their head. The colleague sitting next to them starts from zero every time. There is no shared brain — only a shared invoice.

Prompts are long because context isn’t durable. Every conversation begins with the same preamble: who we are, what our fields mean, how our data interconnects, what we’ve already tried. A huge fraction of every prompt is a re-explanation of the firm to a model that has no memory of the last one. Token waste, sure — but more importantly, attention waste.

Vendor concentration risk is real. A team that built six months of muscle memory on one harness or one model has built it on someone else’s roadmap. If the vendor pivots, raises prices, or goes away, the team’s accumulated effort goes with it. The work was real; the asset was a rental.

A junior cannot pick a senior’s brain on the first prompt. Today, the only way to onboard is to learn what to prompt for, when, and how. That knowledge is tacit and slow. The promise of AI was that institutional knowledge would become queryable. So far, it hasn’t.

A private brain for the department.

The fix is structurally simple, even if the engineering isn’t. The team needs one durable layer that sits between its people, its systems, and whatever model it happens to be using today. That layer holds the context, the prior judgments, the team’s vocabulary, the governance rules, and the memory of every prior interaction. When the team prompts a model, the layer hands it exactly the context it needs — pre-loaded, pre-validated, scoped to the question. When the team gets an answer back, the layer learns from it.

This is what Finny is. The brain a finance team owns, attached to whichever agent or model the team prefers, learning from every close, memo, reconciliation, and review the team ships.

The brain is private. The brain is portable. The brain is the team’s, not the vendor’s.

Read the product ↗

The old agent playbook was built for weaker models. The new one isn’t.

For a long time, “AI agent” meant something narrow: a thin wrapper around a model that could call two or three tools and then hand control back. That design made sense when the models couldn’t be trusted to do more than one step at a time. Those models are gone.

The next step is to let agents operate like interns — capable of real work, accountable to a supervisor, working from a brief, producing artefacts that a human reviews and approves. The shape of the work changes accordingly.

What it takes to build a brain that compounds.

The future of the firm is not a frontier model. It is a frontier ecosystem.

The last thing any of us should want is a world where every company, in every sector, cedes value to a small number of models that absorb everything they see. If all the economic returns of this transition accrue to three or four model providers, the political economy will not tolerate it. There is no societal permission for an AI future that hollows out entire industries.

We have seen this movie before. In the first phase of globalization, the GDP numbers looked fine on the surface while entire industrial ecosystems were being hollowed out by outsourcing. The displacement was real and the consequences are still being felt. Let’s not bring that dynamic into the AI era — a small handful of systems capturing the returns while every industry finds its knowledge commoditized right out from underneath it.

The work, then, is to build a frontier ecosystem rather than just a frontier model. One where every firm, in every industry, in every country, can own the learning loop that encodes its institutional knowledge. One where value flows broadly because every organization can compound its own human and token capital.

This is the ethos we grew up with on the best platforms — where more value is created on top than is captured inside. When that happens, firms create value for themselves and for the economy around them. Employees see their expertise amplified, their judgment becomes part of systems that are replicable and scalable, and the benefits accrue back to the communities those firms sit inside.

“That is the stable equilibrium we should build together.”