No AI is ever going to jail for you

Everyone in accounting is talking about what AI can do. Almost no one's asking what it can't.

I've worked in tech for over twenty years, twelve of them in accounting software, and I've sat through enough hype cycles to know this one is different. So this year I set out to immerse myself in AI and work out which companies in our space are doing something real, rather than slapping a sticker on the website. That research became a report, the A to Z of AI Accounting Software, and I pulled its core ideas into a session at the AI in Practice Summit. You can watch the full talk here.

Here's where I've landed. Most of the AI money in accounting is going into the wrong part of the job. The problems that decide who actually wins are human, and no one's building for them.

The problem AI skipped: sixty days before the ledger knows anything

A conversation with Paul Lauder at Dext gave me a number I'd never heard before. The delay-day index. It's fifty to sixty. That's how many days pass between a document being created, an invoice raised or a transaction happening, and the ledger ever seeing it. Two months before any system of record knows what actually happened. And that number has barely moved in ten years.

Sit with that. We talk endlessly about automation and bank reconciliation, and it still takes two months to get information from a client into the general ledger.

The two human problems AI leaves untouched: apathy and fear

Nearly all the AI money has gone into the middle of the workflow. Classifying, reconciling, reporting. Coding transactions and posting journals. That's useful, but it steps around the two problems that actually matter, and both of them are human.

The first is apathy. "I'm too busy." The shoebox that turns up every quarter. That's the fifty to sixty days, and I think we'll get that number down.

The second is harder. It's fear. The fear of getting things wrong. That's the trust a client places in their accountant, and AI does very little to ease it. If anything, it makes people more nervous, once you add in hallucinations and the idea of letting software do the books on its own.

The real moat is liability, not data or distribution

Hold onto that fear for a second, because it points straight at the most valuable thing an accountant owns. When people talk about AI's impact, they usually name two moats. Data and distribution. The big players sit on vast datasets, and the established names have customers to sell to.

But there's a third one, and it was the single biggest penny-drop for me. Liability. The best line I've heard on it came from Greg Sheehan: people don't use an accountant because they're silly and can't do the sums. They use an accountant to outsource the anxiety of getting things wrong.

That's the whole thing. The business that wins this isn't the one with the cleverest AI or the biggest dataset. It's the one people trust. For AI to replace the accountant, it has to replace that trust. And who, right now, would hand their books to AI on its own? Every piece of research I saw in my years at Xero says the same thing. Accountants are the number-one trusted adviser to small businesses, ahead of lawyers and consultants. That trust takes years to earn, and it's the deepest moat the profession has.

The ledger becomes infrastructure, and 80% of software is exposed

So trust is the human moat, and it holds. The software underneath it is shifting fast. The question I keep hearing is whether AI sits on top of the ledger. I think that's backwards. The real question is whether the ledger sits underneath the AI.

Today, you spend your day inside Xero, QuickBooks or Sage. I bet that tomorrow they slip underneath, into the infrastructure. And the thing you actually work in becomes a plain-language conversation sitting on top. It's happened before. Stripe made payments invisible, a quiet service humming behind every checkout. Xero Force is the clearest signal yet that the incumbents can see it coming, and orchestration tools like Archie are already learning to work across the ledger and the apps around it.

That splits the market in two. Orchestration engines at the top, the big ledgers as infrastructure at the bottom, and a gap in the middle. Around eighty per cent of the software in the ecosystem is built on someone else's pipes. No deterministic data of its own, no orchestration. Just a surface, a nicer way of seeing things. That is a lot of software at real risk, and slapping "AI native" on the website changes none of it. A sticker is not a strategy.

The Great Reset is coming, and it only takes one firm

So when does everyone feel this? I call it the Great Reset, and the only thing I can guarantee is that my date is wrong. My best guess is December 2027, somewhere between September 2027 and June 2028. Building software has already collapsed in cost by orders of magnitude, orchestration is landing now, and one variable matters more than the rest. It only takes one. As Stuart McLeod put it, a single large firm repricing is enough for the whole market to realise the game has changed.

What to do now: adopt the orchestration layer and test on one segment

If you run a practice, start with the orchestration layer and try it on a single client segment this quarter. Test and learn beats reading about it, and it certainly beats arguing about it on LinkedIn.

Capture what I call your wake code, too. The unspoken decisions and judgement that live in your senior people's heads. Getting it written down matters enormously, especially if your senior partners are planning to retire.

One caution. I keep seeing people say they vibe-coded an accounting tool over the weekend on Claude. You can do genuinely useful things that way, but running a practice on top of data that was never built properly is a real risk. The tools made for accounting are building deterministic systems with auditability and traceability, and that is the point. Something like Briefcase is a sensible place to start, because you can train it and correct it rather than letting it loose on your books.

Clients never paid for software. They pay for trust.

I'll leave you with the punchy version. Software as we know it is going away. Clients never paid for clicks, dropdowns and dashboards. They pay for the service, for the person who removes the anxiety and helps them understand their own numbers. The firms that win this wave will be technology-savvy, because they have to be, but they'll win on trust and on genuinely caring about the work.

This is bigger than the cloud. I think it's a seismic shift, bigger than anything we've seen in tech. And nothing coming can replace the trust your clients place in you. That's worth having a front-row seat for. Watch the full session here.