Concepts

The rails beneath the products people actually touch. Most of what matters in finance is invisible: who can move value, who can be trusted with credit, who gets to participate at all. I build at that layer, because whoever owns the infrastructure quietly sets the limits of everything built on top of it.

Rockefeller’s rule: own the rails, and let everyone else fight over the cargo. Real power in any system sits at the bottleneck others must pass through — the refinery, the compute layer, the settlement rail. Visible activity is rarely where the leverage is.

Banks do not disappear; they get absorbed. As banking dissolves into a set of functions any company can call, the regulated core sinks beneath the platforms people already use every day — a grocery app, a telco, a marketplace — until the bank becomes invisible infrastructure inside someone else’s surface. The winner is not the most famous consumer bank but the regulated engine humming behind the distribution. This is the thesis behind ColendiBank and the Migros integration, and it is the spine the rest of these ideas hang from.

In consumer finance, distribution beats brand. A bank spends a decade and a fortune becoming a trusted name; a platform that already owns daily attention can put the same product in front of millions overnight. ColendiBank’s strength is not to become a celebrated consumer marque but to be the regulated engine inside surfaces that already have the customers. The scarce asset is not the logo — it is the place where the customer already is.

The product form absorption takes. Banking delivered natively inside the platforms where people already spend their time — grocery, telecom, commerce, payroll, SME software — instead of through a standalone bank channel nobody chooses to open. Credit at the checkout, an account inside an app opened for something else, payments that vanish into the workflow. The bank stops being a destination and becomes a feature of everywhere else.

For five thousand years money was a noun — something you hold. It is becoming a verb — something that executes. Programmable money carries its own logic: it enforces conditions, settles itself, and operates without you. That is the most significant monetary shift since the invention of credit.

A bank whose operating system is AI. Humans set purpose, risk appetite, and governance; agents execute the bounded workflows — onboarding, underwriting, monitoring, collections, support — that used to consume the organization. The interesting question stops being what the machine can do and becomes what humans must remain accountable for, and where the limits of an agent’s authority are drawn. Not finance with a chatbot bolted on, but the institution rebuilt with agents as the default operator and people as the governors.

The counterweight to the agentic bank. If agents operate the institution, the serious question is not intelligence but control: authority limits, audit trails, evaluations, citations, rollback, and a human who is accountable when it goes wrong. “The model decided” is not a defense a regulator accepts. The control plane is the layer that makes autonomous operation legible, reversible, and owned — the difference between an AI bank you can actually run and one you can only apologize for.

Trust has to live somewhere. For a century it lived in institutions — the branch, the brand, the regulator. It is migrating into protocols, models, and code, and the design question of this era is how you rebuild it there: verifiable identity, explicit permissions, immutable audit trails, and clear accountability for every automated act. Some of trust can move into software; some must stay a human responsibility. Getting that boundary right is the whole game.

Collateral is how the old system priced trust for people it could not see. But real-life behavior — payments, commerce, platform activity, telco signals — describes creditworthiness more richly than a pledged asset ever did, and unlike collateral it is portable across borders and institutions. The bet behind Colendi is that data can substitute for collateral, turning credibility into something a person owns and carries rather than something an institution grants or denies.

To govern something, a state — or a company, or a model — first makes it legible, flattening messy reality into countable categories. The danger, as James Scott showed, is mistaking the simplified map for the territory, then reshaping the territory to fit the map. It is the failure mode lurking in AI, banking, and regulation alike.

An institution that remembers is an institution that compounds. Most organizations forget — decisions, policies, incidents, postmortems, the evidence once shown to a regulator — and re-learn the same lessons at rising cost. A bank built on institutional memory captures all of it as a living, queryable substrate, so each new decision stands on the accumulated judgment of every prior one. This is where Memex, Bank Memex, and agentic operations converge: agents are only as good as the memory they reason over.

Strategy sets direction; tempo decides who wins. In regulated infrastructure the advantage goes to whoever can move through licensing, integration, and shipping faster than incumbents can even perceive the shift — building the option set before the market reprices it. Speed is not recklessness; it is the discipline of compressing the distance between a decision and its consequence. By the time the slow understand what happened, the fast have already changed the board.

Compounding is not only for capital. Books, notes, relationships, decisions, and hard-won operating lessons compound into judgment — if you build a system that keeps them and connects them instead of letting each year start from zero. This is the premise behind my Memex: capture everything, link it, and let understanding accrue interest. The rarest asset a founder owns is not an idea but the compounded judgment behind the next thousand of them.

Every technological revolution runs the same arc — installation, frenzy, crash, deployment, golden age. Bubbles are not noise; they are how new infrastructure gets funded before it is useful. Knowing where you stand in the cycle tells you whether to build, to wait, or to brace.

Civilization is not only invention; it is upkeep. The glamorous part of any system is the launch; the real work is maintenance — reliability, repair, sustainment. In financial infrastructure especially, uptime beats launch theatre, and the institutions that endure treat maintenance as the main event.

The ability of institutions to actually do hard things at scale. The West’s best engineers drifted into building conveniences while ceding serious problems — defense, infrastructure, institutional strength. A live question for me is whether technology returns to serving state-scale problems, and on whose terms.