For three years the AI story in accounting was a story about features. A copilot tucked into the corner of a screen, a smart suggestion in the reconciliation panel, an assistant that drafted a client email. Useful, some of it. But all of it sat on top of software whose core was designed decades ago, back when the job was to record what already happened and close the books once a month.

Walk the floor at AICPA Engage, the profession’s largest US conference, held in early June, and a different pitch had taken over the room. The story was no longer AI added to the ledger. It was the ledger rebuilt around AI. A small group of companies is making the same wager: that you do not bolt intelligence onto a thirty-year-old general ledger, you build a new one with the agents on the inside. The category has a name now, and a surprising amount of money behind it.

The clearest example is Digits, which brands itself as the world’s first “Agentic General Ledger.” Set the trademark aside and the idea is concrete. Instead of an accountant coding transactions while the software offers hints, the ledger itself runs agents that code transactions, reconcile bank feeds, and update schedules continuously, with the human moving to review and approval. Accounting Today named it a 2026 Top New Product for accountants, and Digits has spent the year shipping in that direction, adding an AI-native accrual system in May and, in April, a connection that lets outside AI tools talk directly to its ledger. Its own framing is the tell. The company describes accounting moving from a periodic, month-end event to an always-on process that hums in the background.

Puzzle is making a similar bet, aimed at startups and the firms that serve them. It pulls data straight from the tools a young company already runs, Stripe, Brex, Mercury, Ramp, and turns the feed into reconciled ledger entries and statements. The company says it automates the large majority of the repetitive work behind the books. The thread connecting it to Digits is architecture, not marketing. These are not QuickBooks with a chatbot stapled on. They are ledgers designed, from the first line of code, on the assumption that an agent does the first pass and a person does the judging.

If product claims were the only signal, you could file all of this under hype and move on. The funding is harder to wave away. Digits says it is backed by nearly $100 million from investors including Benchmark, SoftBank, and Google’s venture arm. Puzzle raised another $30 million to keep building. And the same wave is funding a second flavor of the idea, pointed not at the business’s books but at the accounting firm’s production line.

Basis, which builds AI agents for firms across tax, audit, and advisory, raised $100 million in February at a $1.15 billion valuation, led by Accel and Google Ventures with former Goldman Sachs chief Lloyd Blankfein joining the round. The company says its agents already run inside roughly 30 percent of the top 25 US accounting firms, handling document review, reconciliation, and tax preparation while a human stays in the loop to review and sign. Accrual, founded by engineers who built financial infrastructure at Stripe and Brex, came out of stealth the same month with $75 million, most of it from General Catalyst, and a client list that already includes H&R Block and Armanino. Its pitch is to fold tax preparation and review into a single system without surrendering the controls and audit trail a firm depends on.

Put the four together and the shape is clear. One set of companies is rebuilding the ledger the business keeps. The other is rebuilding the assembly line the firm runs on. Both start from the same premise, that the software should produce the first draft and the human should judge it, and both are being funded as if that premise is about to become the default rather than the experiment.

The shift that matters here is not speed. It is rhythm. The monthly close exists because old software recorded transactions and a person periodically went in to categorize, reconcile, and adjust them in a batch. If agents do that work continuously, the close stops being a deadline and becomes a state the books are simply always in. Digits describes precisely this, accounting moving from a periodic event to an always-on process, and it is the logical end point of the architecture. In a first-hand recap of the Engage floor for the site The AI Accountant, Peter McCarroll, a client-accounting consultant who walked it, noted that CPA.com is now floating the “zero-day close,” books that are always current, as an achievable target rather than a fantasy. Treat the timeline with healthy skepticism, because the people selling the tools and the people consulting on them both have reasons to sound urgent. The direction, though, is exactly what these products are built to deliver.

If it holds even halfway, the work does not disappear. It moves up the ladder. The valuable hour stops being the one spent categorizing a thousand transactions and becomes the one spent catching the categorization the agent got confidently wrong. We have made some version of this argument in every edition since the first one. What is new is that the software is now being designed around it. The pitch has quietly changed from “AI helps you close faster” to “AI runs the close, you supervise it.” That is a more honest description of where this goes. It is also a much bigger ask of whoever is doing the supervising.

Now the part the demos skip. Each of these ledgers is its own world. McCarroll went to Engage intending to ask vendors the hard question to their faces, whether their walls are open, and came back with a blunt verdict: nobody on the floor has open walls yet. Puzzle and Digits both publish APIs, he found, but those integrations pull data in without letting an outside system interrogate or instruct the ledger. Data flows in, control does not flow across. The agents are impressive inside their own walls and indifferent to everyone else’s. Adopting one is not installing a plugin next to your existing setup. It is moving your foundation onto someone else’s platform and betting they remain the best home for it for years.

That weighs more here than almost anywhere else in the stack, because the general ledger is the one component you cannot casually swap. For a brand-new firm or a startup with no history, picking an AI-native ledger is a clean decision. For an established practice with years of clients sitting in QuickBooks or Xero, it is a migration, a retraining, and a long-term dependency all at once. None of that is a reason to dismiss the category. It is a reason to read the arrangement the way you would read a commercial lease rather than a free trial. Ask where your data goes if you leave, whether your books export into a form a competing system can actually ingest, and who controls the agents posting entries in your clients’ names.

There is one more question independence forces, because no vendor will raise it for you. Some of these companies sell you the software and also, on another page of the same website, do the accounting themselves. Digits offers both its platform and in-house accounting services. Puzzle has gone the opposite way and states plainly that it will never offer competing bookkeeping services, specifically so the firms building on it are not feeding a future rival. Neither model is wrong, and both can be good products. But if you are going to build your practice on a vendor’s rails, it is fair to know whether that vendor is purely a supplier or also, in another line of business, a competitor for the same clients.

The skeptical read can be overdone, and the other side deserves a real hearing. Real-time books beat books that are six weeks stale, every time. A firm drowning in low-value compliance work and unable to hire its way out, the exact squeeze we covered in our talent edition, has a genuine reason to want agents doing the first pass so its people can spend their hours on judgment. A junior who spends year one interrogating an agent’s output, asking why it coded something a certain way and where it is most likely wrong, may build real judgment faster than one who spent that year keying in receipts. And the early movers, on the numbers the vendors share, are taking on more client work without adding staff in lockstep. If you are starting a practice today with no legacy system to defend, building it AI-native from day one is a defensible bet. It might be the obvious one.

So what do you do with this. Not panic, and not tear out a working system on Monday morning. The honest near-term move is smaller and more useful than either.

Treat the category as real and structural rather than a passing booth at a trade show. Remember too that it is early: at Engage, Puzzle and Digits both told McCarroll plainly they are not yet ready for the Canadian market, with multi-currency and sales-tax handling still missing, a useful reminder that “AI-native” does not yet mean complete for every jurisdiction. Pilot one of these tools on a slice of the work, a handful of clients or your own internal books, before you believe any percentage printed on a slide. When you pilot, judge it as a platform decision and not a feature purchase: data portability, whether it competes with you, what your review workflow actually becomes, and what breaks if the vendor lifts its prices or gets acquired. Keep a human accountable for every entry that carries your name, which is the same discipline we have preached since edition one and which matters more, not less, once the entries are being generated around the clock. And take the one piece of conference advice that travels well: for the core systems, the ledger, the tax engine, payroll, you buy rather than build, so the only real question is whose ledger, and on what terms.

The skill that gains value through all of this is not knowing where the buttons are. It is being the person who can look at an always-on, agent-run set of books and tell, fast, where they are wrong. That person was always the actual job. The software is finally being built to admit it.

The last wave of AI in accounting was a fresh coat of paint on old software. This one is a new foundation, and the funding, the products, and the conference floor are all pointing the same direction. The general ledger is being rebuilt for agents, and the close is turning from a monthly scramble into a continuous state. That is a real shift, and for some firms a genuinely good one. Just be clear about what you are choosing when you choose one. You are not buying a feature. You are picking the ground your practice will stand on, and the walls that come with it.

— Footnote

Footnote is an independent publication, with no affiliate links and no vendor paying for placement. We have no commercial relationship with any company named here. This is informational, not professional accounting, tax, or legal advice. Funding figures, customer counts, and automation rates come from the companies’ own announcements and the reporting linked above, are current as of June 2026, and in several places reflect vendor claims we have attributed rather than independently tested.