Credit at machine speed.

About 1.4 billion adults still have no account at a bank or a mobile money provider. Small businesses in developing economies face a financing gap estimated in the trillions of dollars. The usual explanation is risk. We think the usual explanation is wrong.

A loan is priced by what it costs to understand the borrower. A $3,000 loan to a fisherman takes the same analyst, the same calls, and the same file as a $300,000 loan to a company with audited books. On the small loan, that work eats the margin. On the unusual one, there is no model to lean on. So the loan is never declined. It is never priced.

The market that results is not small. It is the part of the world that lending skipped because reading it was too expensive.

Every unique business case is a research problem: what the business does, who pays it, when, and what could stop the money. People do this research well, slowly, and at a cost. They also get tired of edge cases, and edge cases are most of the world.

Agents do the same research in parallel and in minutes. They read everything, they ask the follow-up question, and the hundredth strange business gets the same attention as the first. The cost of understanding a borrower falls toward zero. When it does, the loans that never made economic sense start to.

Antal is developing frontier models for credit. They are not a score with a chat window on top. They answer three questions a lender has always had to answer by hand, for businesses no rate sheet was written for.

1. 01

   Research

   What is this business, really?

   Deep research on a company from the work itself: what it sells, who pays it, how often, and what the season does to the cash. The conversations an owner chooses to connect are part of the record.

2. 02

   Match

   Who should lend, and how much?

   Finds the right capital for a case no rate sheet was written for, and sets a limit the business can carry instead of a limit the form allows.

3. 03

   Repay

   How does the money come back?

   Shapes repayment to how the business is actually paid. A fisherman paid weekly repays weekly. A shop paid at the register repays from the register.

The models learn from our own lending book. Every line we extend, every payment, and every miss goes back into the next decision. That is why we lend and build models in the same company: the book is the training data, and the models are how the book grows.

The agent is already where the company's work happens. It writes the code, answers the customers, books the jobs, and buys the tools. It sees the business more clearly than any form an owner fills in once a year.

It can also be bounded in ways a person can't. A card held by an agent has a limit the agent cannot raise, a number the agent never sees in full, and an owner who confirms anything that changes the company. The card is the first credit product an agent can hold safely. It won't be the last.

Limits grow the way trust grows: with evidence. Each layer below is something we can verify, price, and insure.

1. 01

   The work

   Connected conversations and the company's activity set the first limit.

2. 02

   A certified agent

   We certify and insure the agent that runs the card, so the agent itself becomes underwritable.

3. 03

   Personal Guarantee Insurance

   Insuring the owner's guarantee unlocks a line of credit above $100,000.

Today a borrower fills in a form and a person reads it. Tomorrow the company's agent asks, and Antal's agent underwrites, prices, and services the line. The question and the answer move at the speed of the work.

Lending stops being a desk. It becomes a conversation between two agents, with a person confirming the decisions that matter.

Better models find borrowers no one else can price. Those loans grow the book. The book's outcomes train the models. Each repaid loan makes the next decision cheaper and more accurate, and each new kind of business teaches the models something no competitor has seen.

1. Stage one

   The agent card

   The first AI agent credit card. Only the agent manages it. The limit comes from the work.

2. Stage two

   Lines on what the company owns

   Credit against property, equipment, and revenue, underwritten by the same models.

3. Stage three

   The loans no desk would take

   Small, unusual businesses anywhere, underwritten and serviced agent to agent.

Credit at machine speed fails at machine speed too. We design for that from the start.

Fraud

An agent can fabricate a story faster than a person can. Every limit is tied to evidence we can verify, not to what an agent claims.

Misuse

The agent never sees the full card number, and anything that changes the company comes back to a person as a confirmation link.

Bad models

Limits start small and grow with repayment, so a wrong answer costs little and teaches the model something.

Regulation

Credit is regulated for good reasons. We work inside those rules and with licensed partners where they require it.