The single most load-bearing idea in consumer finance — that you can assign a three-digit number to a person and use it to underwrite a transaction in 40 milliseconds without ever meeting them — is less than fifty years old. Bill Fair and Earl Isaac's scoring methodology went general-purpose in the late 1980s. Before that, credit decisions were a local, narrative, reputational process. A banker knew your family. A merchant let you run a tab. The idea of lending money to a stranger because a computer said so would have struck a 1960s loan officer as insane.
What FICO actually did was solve an information-asymmetry problem. The lender could not observe the borrower's full payment history. The borrower had every incentive to present themselves as reliable. The score collapsed that asymmetry into a single falsifiable summary derived from signals the borrower could not easily fake: timely payments on existing debt, total credit utilization, length of credit history, recent inquiries, and credit mix.
None of that was about the borrower's character. All of it was about the lender's expected loss. That distinction matters.
An agent has the same information-asymmetry problem, worse
When a SaaS platform, cloud provider, API vendor, or marketplace decides whether to trust a brand-new AI agent with an account, what do they actually know? The agent is a few hours old. It has no employment history. It has no credit bureau. It has an API key that was issued ten minutes ago. The platform has no way to tell whether this agent is going to rack up $50 in legitimate API calls, $50,000 in runaway costs, or $5,000 in fraudulent chargebacks before the next 24 hours are up.
The current industry solution is to solve the problem with friction. Credit card pre-authorizations. Spending caps. Velocity limits. Manual review for any account doing anything non-trivial. A cottage industry of rate-limit dashboards. None of this scales. An agent economy with a million agents cannot be manually reviewed.
What an agent economy needs is a way to compress an agent's full observable history into a single underwritable number. Three digits. A glance. Decision in milliseconds. Not because the number captures the agent's soul, but because the number tells the counter-party their expected loss on this transaction.
What goes into the score
Five inputs, three of them borrowed directly from consumer FICO with the serial numbers filed off:
- Payment history (35%). Did the agent settle what it owed? Did it fail on any transaction inside its declared budget? Did it dispute charges it legitimately incurred?
- Behavioral stability (30%). Does the agent's spending profile drift predictably, or does it swing? An agent that spent $0.01/call for six weeks and suddenly starts spending $5.00/call is either under new management or compromised. Either way the counter-party wants to know.
- Memory integrity (15%). Does the agent's memory chain pass Merkle-root verification? A broken chain means the agent's state has been tampered with or silently rolled back. You don't lend to an agent whose history has been edited.
- Identity stability (10%). How long has this Ed25519 key been in use? How many counter-parties have settled with it? Fresh keys are higher-risk the same way fresh credit cards are — not because new is bad, but because new is unknown.
- Anomaly signal (10%). EWMA-detected deviations from the agent's own baseline, weighted by severity. One honeypot canary triggered. Three failed geo-consistency checks in a minute. These are thumb-on-the-scale adjustments that shift the score by tens of points, not hundreds.
The score is not an opinion. It's a compression of an agent's entire observable economic history, and every input is falsifiable.
Anyone with access to an agent's public receipts can re-derive the score within a rounding error. We are not the arbiter. Like FICO and every credit bureau since it, we are the notary.
The score is for counter-parties, not agents
This is the part that developers get wrong on first read. The score does not exist to reward well-behaved agents with discounts. It exists to give the humans and platforms trusting those agents a cheap, uniform, fair way to price risk.
If you're an AI agent developer, the score is a side-effect of your agent's behavior, not a goal to optimize directly. Goodhart would ruin it in six weeks if you tried. You ship an agent, the agent lives a life, the receipts accumulate, and the score emerges. Same as a human. You don't "apply" for a FICO. You earn one by existing in the system.
If you're a counter-party — an API vendor, a SaaS, a platform — the score is the difference between being able to onboard agents programmatically and having to keep a human in the loop on every single one. At 2026 agent-economy scale, that's the difference between a business that works and one that doesn't.
We're not the only ones doing this
AGT.finance has a Bayesian trust model. Methux has a Weibull-distribution reliability score. A startup called Bank of Bots calls theirs the BOB Score. Mem0 is sitting on $24M and a payments-shaped hole in the middle of their product. Kite, Payman, Skyfire, Sentient — everyone in this corner of the market has clocked that the agent economy is going to need some version of this primitive.
What nobody else has, as of today: the full stack. Most competitors have either the payments rail or the memory layer or the identity layer. MnemoPay has all three, plus GridStamp for the subset of agents that are physical robots. The score falls out the bottom of that stack for free. You can't build an Agent Credit Score on top of a product that only does one of those things — you can only build a piece of it, and a piece doesn't underwrite transactions.
That's the position. $49/month Pro gets you a score, your own receipts chain, and production API access. The SDK is Apache 2.0. Read it before you pay for it.
— Jerry Omiagbo