Signature economy
Somewhere right now, a controller is running a nine-agent close pipeline she assembled in Lovable. A CFO has a close command center living in Claude — projected close duration, a waterfall of where the days get lost, exceptions ranked by exposure. Another is running ten thousand Monte Carlo simulations on next quarter's EBITDA in a Colab notebook an AI wrote for him.
None of these people write code. All of them shipped software this year. It's one of the most impressive things happening in finance right now, and it deserves more credit than it gets.
Spend enough time with finance leaders who use AI every day and you start to hear the same three wishes, almost word for word.
What they're asking for
When month-end hits, I want reconciliation across bank, AP, AR and GL to happen automatically - so close takes days, not weeks, and the numbers that flow to the board are defensible.
The close. Reconciliation is the single largest manual time sink in the cycle. ERPs do the bookkeeping well, but the specifics - multi-source matching, multi-entity, FX, revenue recognition - tend to break the box. The dedicated tools that solve this, BlackLine and Trintech and Workiva, are excellent at what they do, but they're built and priced for the enterprise. So controllers reconcile in spreadsheets. And now, sensibly, they ask agents to help.
When the business is off plan, I want to model what's next and explain why we're here - so leadership has analysis to act on, not numbers to react to.
Modeling is where most of the AI effort goes - scenarios in minutes, variance explained by drivers, profitability sliced by product and customer. It's also where the ceiling shows first: even a good model can't explain a variance when the data underneath is messy.
When the quarterly cycle hits, I want the board pack to assemble itself from the five systems where the numbers live - so I deliver an accurate, on-time view without losing two days to stitching.
And the pack - still hand-assembled every cycle, even at companies that own Oracle Hyperion. The deck itself is largely solved; AI writes good slides now. The numbers going into it are the hard part.
Three wishes, and on a closer look they're not really three. The analysis stands on the close. The pack stands on the analysis. The close is where the numbers get made - improve it, and everything downstream inherits the improvement.
Why even the best prototypes stall
What struck us most in all of this wasn't the ambition, though there is plenty. It was how consistently the same honest answer came back when we asked why a working prototype hadn't made it into a real close:
I can't verify what the AI did
There's no audit trail
I can't stand behind the number
That last sentence, we've come to believe, is the heart of it.
Finance is a signature economy. The product of a finance team isn't analysis - it's a number someone is personally willing to sign. When a controller signs the close, they underwrite it with their name. When a CFO takes the pack to the board, every figure in it is a personal representation. And behind every signature stands the profession's quiet end-user: the auditor, who arrives months later and asks to see how the number was made. Signatures carry liability; liability needs evidence; evidence is a trail. A tool that produces the right answer but can't show its work isn't yet a tool this profession can use - however good the output, it remains a liability with good UX.
Seen this way, the hesitation looks less like resistance to change and more like professionalism. The push is real - many describe falling behind on AI as a career risk, and the grind of close and board pack as the work that hurts most. The pull is real too - visible productivity gains, peers sharing genuinely useful pipelines. What sits in between isn't fear of the new. It's the entirely reasonable refusal to sign what you can't inspect. Even the most technical people in this market say they'd need an engineer to keep anything running in production - and even then, the deeper issue remains: the trail was never there by construction, and no prompt can retrofit it.
So finance teams today sit between two imperfect options: AI tools they can build but can't yet defend, and enterprise systems they can defend but can't easily justify - in price or in rollout time. One takes a weekend and stops at the demo. The other takes a year and a budget line.
What the trail is actually worth
We've seen what lives inside that gap. Earlier this year we connected an agentic ledger to a national commercial general contractor running a twenty-year-old ERP, and reconciled its job costing against three years of audited financials. The team there was diligent, and the books were clean by every normal standard - they had passed every audit. The reconciliation still surfaced commitments that had been quietly relieved with no matching invoice, purchase orders still holding budget on jobs closed years earlier, and roughly $180,000 of recoverable margin sitting in the gaps between systems. No one had done anything wrong. The systems had simply never been asked to show their work - and money tends to hide wherever work goes unshown.
For us, that was the quiet lesson. The audit trail isn't compliance overhead, a tax paid to keep the auditor calm. Provenance is how errors get found, and errors are where margin leaks. The trail isn't the cost of the system. It's the yield.
Accounting is the control plane
Lately the industry has a name for what it's searching for: a control plane. Control planes for agent payments, for treasury, for AI spend - a new one every week. The word is right; the implementations are thin. Spending caps and approval buttons aren't a control plane - they're a fuse box. The control plane finance teams already trust has existed for five centuries, and it's called accounting: double-entry, the period close, the audit trail - the system through which money is understood and verified before it moves. So the real work isn't bolting controls beside the ledger, or bolting AI on top of it. It's rebuilding the accounting system itself so agents can operate inside it - closing the books, explaining variance, triaging exceptions, and, in time, moving money under the same discipline that records it. The agents and the accounting have to be designed together.
There is a longer arc here, too. We've argued that money is becoming code and that agents will be the next customers of banks. Both of those futures quietly assume a missing piece: before an agent can be trusted to move money, someone has to be able to stand behind the numbers it computed. Signatures come before payments. A verifiable ledger isn't a detour from that thesis - it's the trust layer the rest of it stands on, and the reason it comes first.
The missing block
What's missing, we think, isn't a better copilot or another dashboard. It's a ledger built for agent work - a system of record where verification is the substrate, not a feature.
Concretely: every ingestion, match, and adjustment logged with provenance. Any number drills down to its source document. Deterministic where determinism matters - debits, credits, eliminations - and probabilistic where judgment lives, with the boundary between the two kept visible. An audit trail that exists because of how the system works, not as an export bolted on afterwards.
And it shouldn't be another destination. Finance work already lives in spreadsheets, chat, and AI assistants - we made this argument in The bank with no interface. The ledger belongs underneath: exposed as tools and skills to Claude, ChatGPT and Copilot, with its own surface only for the moments that need one. It sits alongside the ERP rather than replacing it - the ERP keeps the books; the ledger makes the work defensible. And it should be priced for the SMB and mid-market teams doing all of this by hand today, not only for the enterprises that already bought BlackLine.
Finance software used to compete on interface, then on intelligence. A CFO can now generate a dashboard before lunch and a scenario model after it - those moats have quietly dissolved. What no prompt can generate is the right to stand behind the output. Defensible turns out to be the operative word twice over: the number a CFO can defend, and the business built on making that possible.
This market has already proven the demand in the most credible way a market can: by building the tools itself, on weekends, with whatever was at hand. It isn't waiting for capability. It's waiting for the block that lets it sign.