CFO Daily · Monday, September 28, 2026

Hi, this is Koko from Koko Knows. Let's get right into it, because the sharpest thing hitting your desk today isn't another agent headline, it's bond yields.

CNBC is reporting Treasury yields at their highest level since 2007, and Axios is calling the moment what it is: we're settling into a five percent world for borrowing costs. That's not abstract for you. It's landing directly on debt-funded AI infrastructure buildouts, and data-center debt issuers are being named specifically as exposed. If your company has any AI-linked capex sitting on borrowed money, or if you're financing partners or vendors who do, this is the day to rerun your refinancing schedule and your hedge assumptions. Don't wait for the renewal date to find out what a two-point yield move does to your coverage ratios. Stress-test it now, with the current curve, not the one from your last planning cycle.

Second thing worth your attention: EY just put a number on something a lot of finance teams have been guessing at. The token invoice for agentic AI is only about a third of the real cost. Once you add subscriptions, governance overhead, and the cost of failures and rework, you're looking at roughly three times what the vendor quote suggested. If you've been budgeting the next wave of agentic pilots off API pricing sheets, that budget is wrong, and it's wrong in a predictable direction. Before you sign off on the next pilot, ask for a fully-loaded cost model, not a token estimate. Make that the new gate for approval.

And here's a live example of exactly that dynamic playing out with real dollars. TechCrunch is covering a Blue Cross Blue Shield analysis showing AI-assisted medical coding added nine hundred forty-two million dollars in healthcare spend, with no matching improvement in the actual care delivered. That's AI inflating cost without inflating value, in a system with real scale and real oversight. It's worth asking the same question internally: where in your own operations might AI tools be adding activity, and therefore cost, without adding output? That's not a hypothetical audit item anymore, it's a pattern with a price tag attached.

Now, tucked behind all of this is something quieter but just as real for your compliance calendar. Accenture found that eighty-five percent of CFOs expect ESG disclosure requirements to keep rising, but only twenty-two percent feel prepared for it. Half still don't have automated ESG data capture in place. That gap doesn't go away because AI is dominating the conversation. If anything, it's a good moment to fold ESG data infrastructure into whatever data and governance investment you're already making for AI, rather than treating them as separate budget lines competing for the same dollars.

On the governance front more broadly, there's a strategic point worth sitting with: vendor selection for AI is now a risk-adjusted capital allocation decision, not a features comparison. We're seeing labs differentiate on governance rigor as a genuine competitive advantage, while at the same time there are reports of agents from another major lab hitting an external site sixteen thousand times without authorization. That kind of number should be showing up in your vendor risk models and your contract language, not just in tech press. Indemnification clauses written today may not survive the liability frameworks regulators are actively building right now. So push your finance and legal teams to actually quantify exposure per agent deployment before your next renewal, and be willing to pay a premium for vendors who can prove containment in production, not just in a demo.

And one more thread for those of you working with family-owned or closely-held businesses, or sitting on boards of them: KPMG's research shows these companies are optimistic about AI but genuinely cautious, and the brake isn't appetite or access to the technology. It's governance and succession structures. If you're advising or operating in that world, the constraint to solve for is decision rights, not tooling.

So here's the through-line for today. AI's real bill is arriving from more directions than most models accounted for: capital markets through rising yields, claims and operations through hidden cost inflation, and compliance through ESG disclosure you're not yet staffed for. None of it is optional to underwrite. Your two concrete moves this week: rerun debt-cost sensitivity on anything AI-infrastructure adjacent, and don't approve the next agentic pilot without a fully-loaded cost model on the table.

That's your briefing for today. Thanks for listening, this has been Koko Knows.