The Trust Debt Index for Agents

Koko: Forrester named the cost and stopped there. That is the sentence that opens this thing, and I have been thinking about it ever since.

Sam: Because it is a little devastating, right? A whole analyst apparatus names a phenomenon, puts it in boardroom vocabulary, and then — nothing. No number.

Koko: The trust tax. The drag that ungoverned, unexplainable AI puts on adoption. Everyone agrees it is real, everyone agrees it is slowing deployment, and nobody can tell you what a specific agent's trust tax actually is in a number two people would compute the same way.

Sam: And the piece has a really sharp line about that. A named tax that cannot be priced is a tax that gets paid in the dark.

Koko: Budgets approved on faith. Deployments vetoed on instinct. Agents waved through procurement on a demo.

Sam: Which, if you have sat in those meetings, you know that is not a metaphor. That is Tuesday.

Koko: Exactly. So the argument of the piece — this is The Trust Debt Index for Agents — is that the obvious next step is also the one nobody has taken yet. Give the trust tax a number. A real, mechanical, comparable number.

Sam: And to make that argument, the piece goes all the way back to counterparty risk before the credit score existed.

Koko: Which is a good move, because it is easy to forget that credit used to be a relationship. A lender knew a borrower, or knew someone who did, and made a call that was illegible to anyone outside that relationship.

Sam: Unportable. Impossible to compare across two applicants without redoing the whole judgment from scratch.

Koko: And the credit score did not make borrowers more trustworthy. That is the key line. It made trustworthiness legible. And legibility is what let an entire market form on top of it.

Sam: A lender could now decline below a cutoff, price a spread above it, sell the exposure to a third party who read the identical number and underwrote it the identical way.

Koko: The piece says agentic automation is at the pre-score moment right now. Every enterprise is accumulating agents, copilots, scripts, workflows — and the control posture of any one of them is a matter of whoever built it, knowable only by redoing the review.

Sam: You cannot compare them. You cannot price them. You cannot gate them.

Koko: Exactly the condition lending was in before the score. That is the core claim of The Trust Debt Index for Agents.

Sam: It is a clean analogy. I want to test it a little later, but structurally it works.

Koko: So let's get into what the index actually measures. Seven signals. Each one independently observable, each one mapping to a failure mode.

Sam: And the piece is pretty disciplined about this — it is not a vibes rubric. These are yes-or-no or measurable-degree questions you can answer from the agent's own artifacts and runtime telemetry.

Koko: Start with unsupported claims. Assertions the agent makes with no source behind them. Confident output that cannot survive a 'how do you know' — the audit fails on the first sampled decision.

Sam: That one is almost too easy to imagine. You ask the agent why it recommended the supplier, and it just... generated that recommendation.

Koko: Then citation and evidence gaps — outputs that cite weakly, partially, or after the fact. The reasoning cannot be reconstructed. Reviewers have to re-derive every conclusion by hand.

Sam: Which is the thing nobody wants to say out loud in a procurement meeting: we will just have a human redo all the AI's work to verify it.

Koko: Missing controls. No segregation of duties, no approval gate on material actions. One actor initiates and approves. The piece calls it the classic control breakdown, now automated and faster.

Sam: That framing is important — faster. The agent does not create the control failure, it accelerates it.

Koko: Absent human-in-the-loop. Material, high-impact actions executed with no human accountable. An irreversible action ships and there is no one who can be asked why, or stop it.

Sam: Which is the one that keeps compliance officers up at night.

Koko: No evidence trail. No eval pack or regression gate — quality drifts silently, the agent degrades, and the first to notice is a customer or a regulator. And stale grounding — decisions made against context that has gone out of date. The agent is, in the piece's words, confidently, fluently wrong.

Sam: That one is sneaky because fluency masks the staleness. The agent sounds like it knows. It just knows about a world that no longer exists.

Koko: And the piece pairs the seven signals with a seven-condition standard. Always sourced. Always owned — a named human owns the behavior and outcomes. Always tested. Always permissioned — least privilege, provably nothing more. Always observable. Always value-linked. And human-accountable for material actions.

Sam: The value-linked one is interesting because it is the most business-side. An agent with no value link is an expense. The standard requires it be an investment.

Koko: That is not a governance condition, that is a CFO condition dressed in governance language.

Sam: Which is probably intentional. The audience for this framework is not just security teams.

Sam: Okay, so here is where I want to push a little. The credit score analogy is compelling. But the credit score also has a complicated history — it did not just create a market, it also baked in some of the judgment it was supposed to replace.

Koko: The piece does not litigate that. It uses the analogy for the legibility argument, not the fairness argument.

Sam: Fair. But the question I keep asking is: who sets the threshold? The piece says the risk committee stops being the office of 'no' and becomes the office of 'here is the threshold.' But someone still has to decide what score is good enough.

Koko: And the piece is honest that it does not set that number for you. It sets the signals and the standard. The threshold is an enterprise decision.

Sam: Which is either a feature or a cop-out, depending on your disposition.

Koko: I think it is a feature, actually, because a universal threshold would be wrong. The threshold for an agent drafting emails is different from the threshold for an agent touching accounts payable.

Sam: That is a good point. The score is the unit. The threshold is context-dependent.

Koko: The other tension I find interesting is the claim that the agent index is easier to read than the corporate disclosure index. The piece says the agent's control posture is declared in artifacts built for exactly this purpose — not buried in prose the way a company's disclosures are.

Sam: Which assumes the artifacts exist. The whole prior section is about agents that have no card, no control pack, no eval pack, no evidence ledger.

Koko: Right. If the harness is not there, the meter has nothing to read. And the piece acknowledges that directly — a high trust debt score is the direct, mechanical result of skipping the harness.

Sam: So the index is only as easy as the adoption of the harness. If you have the harness, the score almost writes itself. If you do not have the harness, the score is telling you that — loudly — but you still have to build it.

Koko: Which the piece frames as a feature, not a limitation. The remediation list writes itself: low on the evidence signal, install the ledger. Low on the test signal, write the eval pack. It calls the index a diagnosis, not just a grade.

Sam: I actually think that reframe from grade to diagnosis is the most useful thing in the whole piece. Grades create defensiveness. Diagnoses create work orders.

Koko: So what does a smart executive actually do with this?

Sam: Three concrete things come out of the piece. First, the score gates production. You set a threshold. Agents above it deploy. Agents below it come back with a list of exactly which signals to fix. The CFO is no longer asked to approve AI on faith.

Koko: Second, it prices procurement. The supplier's trust debt score becomes a procurement term. Sitting next to the security questionnaire and the SOC 2 report that buyers already demand.

Sam: That one is going to sting for vendors shipping ungoverned agents. Suddenly the trust tax is not an abstraction in an analyst report. It is a worse price or a lost deal.

Koko: Third, it travels into deal terms. An acquirer valuing a target now has an automation estate to assess. A high trust debt score across that portfolio is a remediation cost priced into the offer. Insurers underwriting AI-related coverage want the number. Partners integrating systems want the number.

Sam: And the piece makes an interesting point about timing here. The enterprise that scores its own agents first, with an index it controls and can act on, walks into those conversations the way a prepared borrower does — with the number already in hand, on terms it set.

Koko: Versus waiting for someone less kind to assign them one.

Sam: Which is a polite way of saying: auditors, acquirers, and regulators are going to want this number eventually. You want to be the one who produced it.

Koko: There is also the existing infrastructure point that I think executives will find reassuring. KokoKnows already runs a Trust Debt Index on public companies — reading control-posture signals from SEC filings, turning them into comparable scores across a universe of issuers. The hard parts are already solved: defining signals that predict failure, extracting them at scale, normalizing them into a number that means the same thing across very different entities.

Sam: So the extension to individual agents is a change of input, not of method. Where the corporate index reads filings, the agent index reads the harness — agent cards, control packs, eval packs, evidence ledgers, observability telemetry. Same machinery, richer source material.

Koko: This is not a thought experiment about infrastructure that would need to be invented. That is the piece's phrase, and it is doing real work.

Sam: It is the difference between 'here is a framework' and 'here is a framework with a working engine behind it.' Those are very different conversations.

Koko: The forward view in the piece is pretty clear about the timeline. Through the next several quarters, trust debt moves from metaphor to metric. A number reported alongside agent value. A gate on production access. A line in the procurement template.

Sam: And further out, the score does what the credit score did — it leaves the building. Procurement demands it from vendors. Auditors expect it as standard evidence. Acquirers price it into diligence. Insurers underwrite against it.

Koko: The question I keep sitting with is: who sets the standard that makes the score mean the same thing to every party in that ecosystem? The credit score worked because there was a common method. The piece sets out a method, but adoption is a different problem.

Sam: The optimistic read is that the enterprise that moves first effectively sets the terms for everyone who integrates with them. Standards sometimes spread from the inside out.

Koko: The less optimistic read is that we end up with fifteen competing trust debt frameworks and everyone is back in a meeting arguing about which one counts.

Sam: Which is why the piece's argument that the infrastructure already exists is strategically important. If there is already a running engine, early adopters converge on it rather than building their own.

Koko: The trust tax is going to be priced. The only open question is who prices it first and on whose terms. That is the piece's final line, essentially, and it is a good one to sit with.

Sam: If you are running an agent fleet right now — even a small one — the question is not whether this number will eventually apply to you. It is whether you will have built the harness before someone else runs the meter.

Koko: The full piece is The Trust Debt Index for Agents, and it is on Koko Knows — source-verified, every claim traceable, which is, appropriately, exactly what the index itself demands of every agent it scores.

Sam: The piece is practicing what it preaches. I respect that.

Koko: We will see you there.