No AI Is Ever Going to Jail for You

Damon Anderson — Founder & Author, A to Z of AI Accounting Software

04:19Why this talk exists

It’s good to be here. I’m going to put my screen up now and jump on full screen — which means I can’t see the questions, so jump in if anything important comes through. I coded this presentation in Claude, and it hasn’t quite got the facility to avoid full-screen mode without the bar across the top. That’ll be in the next round of iterations.

Today’s session is fundamentally about a report I’ve written that sits on a website called the A to Z of AI Accounting Software. As the title says: no AI is ever going to jail for you. I think that’s a good summary of the state of AI in accounting right now. There is a lot of noise — a huge amount of noise — and I set out at the start of this year, as a personal development opportunity, to really immerse myself in the world of AI.

I’ve worked in tech for over twenty years and been through a few hype cycles, but this is one I think is fundamentally different. I spent twelve of those years in the accounting software space, and I decided to jump out of being an operator and build my own business. Part of that is a research arm, which started as me trying to work out which companies are genuinely interesting when it comes to AI. Obviously you’ve got Anthropic and OpenAI, but who inside the accounting software space is doing anything interesting?

06:44Building the report — and vibe coding it in Claude

I started in a spreadsheet and built it out, and then I thought this might be interesting for everyone else, because I’m not the only one who’s confused by every company slapping an AI sticker on its website. So I built the A to Z of AI Accounting Software.

This talk walks through some of the core themes in the first thesis that grounds the report. I’ve vibe coded this presentation in Claude, and it connects with the actual report itself — there are live connections into the report and some of the features on the website, which was also built and developed to start with in Claude. Hopefully it gives you a perspective on who the players are, what I think the market looks like, and some inspiration around what can be created using LLMs to build things.

07:31The truth serum: the delay-day index

Let me start with a truth serum. It came from a great conversation I had with Paul Lauder at Dext, where he told me about the delay-day index — I’d never heard of it. The number is fifty to sixty. That’s the number of days between when a document is created — an invoice is raised or a transaction happens — and when a ledger ever sees it. That’s two months before any system of record knows what happened in reality. And that number hasn’t changed much over the last ten years.

To me that’s a very interesting sign. We talk about AI and accounting, all the automation and bank reconciliation, but it takes sixty days to get the information from a client into the general ledger. This is one of two major areas AI hasn’t really focused on.

08:21The two human problems AI doesn’t solve

The middle of the process — classifying, reconciliation, reporting — is where all the automation and all the AI money has gone. When I look at the landscape, it’s all “how do we categorise and classify and code transactions and post journals?” It’s not solving the client behaviour at the front, which is a truly human behaviour: apathy, “I’m too busy,” procrastination. That’s when the shoebox turns up every month or every quarter.

The other big human trait AI doesn’t solve is fear — the fear of getting things wrong. That’s the trust clients place in accounting and in their accountant. That’s staying out of jail, and that’s unmoved. AI doesn’t do much to solve the fear element. In fact, it probably makes people more worried, because of hallucinations and the idea that you can use AI to do your books automatically.

So we have two areas to think about. The apathy gate — getting those fifty to sixty days down — I think that’s going to happen. But the fear element is more complicated.

10:48Does the ledger sit underneath the AI?

A big question that comes up is: does AI sit on top of the ledger? I think that’s the wrong question. The real question is: does the ledger sit underneath the AI? A lot of the AI we’re seeing is just an augmented chatbot Copilot. What AI is capable of is far more than a general ledger, because a general ledger is fundamentally just a database.

This is the actual answer from the report — there’s a chat on the website, an “ask anything” feature, that uses the report as an LLM to answer questions. The report’s answer is yes, but there’s an inversion happening.

Today, the general ledger is where the accountant spends their day. You spend your day in Xero, in QuickBooks, in Sage. Tomorrow, my view is it becomes infrastructure that an agent is accountable to. There’s a natural-language conversation as the new surface that sits over the top. Xero Force is a clear signal of that from one of the incumbents, but there’s a new layer of orchestration tools sitting over the top that are, in some ways, replacing the place people go on a daily basis.

12:57The big GLs become invisible infrastructure

In short, today’s skyline — the big players you see at the conference stands and use every day — are at risk of becoming infrastructure. That’s not necessarily a bad thing. I’ve seen it happen before, with payments. Payments providers sank into the background. Stripe is invisible; it’s a service to e-commerce vendors.

I think the report is seeing a real trend towards Xero, QuickBooks, and Sage applying that same approach. You see Sage powering a lot of the banking apps right now with Sage accounting; Xero powering Claude; QuickBooks powering Claude. A lot of the work can be done in Claude and the app, but the data — the deterministic layer — sits underneath.

Let me click on Archie as an example. This is a page in the report. Every vendor has an editorial perspective written by me, plus a summary of public use cases, key modules, workflow stages covered, launch status, who the CEO is. The ones at the top are the most interesting because they’re what I call orchestration engines. They sit across the GLs and can orchestrate work inside Xero and the apps around it.

14:32Xero enters the orchestration game

As the report live-answered for me, Xero has entered the game. When I first wrote the report I said they had twelve months to make a strategic call. I thought the Anthropic partnership was reactive and I wasn’t sure they’d be brave enough to rebuild. But in not even twelve weeks, they launched Xero Force — an orchestration hub and core financial operating system. It’s only a few weeks old, so it’s very early to say how it’ll work, but it’s really interesting.

Think about the power Xero has — and QuickBooks and Sage — but I think Xero has nailed the accounting ecosystem. They’ve got an ecosystem of over a thousand apps. If they work out how to orchestrate work across all of those and across Xero itself, they’re in a really strong position. They’ve got years of data to learn from, and that orchestration will let them agentically do work — and build a lot more trust in the work that’s done from a software perspective.

16:48The bifurcation — and the 80% caught in the middle

There’s a real bifurcation. On one side, orchestration engines thinking agentically about how to get work done with other tools, system-agnostic — Archie can plug into Xero, into Dext, into all these things and do work. On the other side, the big GLs becoming infrastructure: solid, deterministic datasets that support agentic workflows.

But there’s a massive gap in the middle. Roughly eighty percent of the software in and around the ecosystem is built on someone else’s pipes. They don’t have the underlying deterministic data that gives them a moat in data and scale, and they don’t have the orchestration capabilities at the top. They’re just the surface — a dashboard, a different way of seeing things. They’re at risk. That’s a huge risk for software providers in that space.

17:32A sticker is not a strategy

It hasn’t stopped that software from saying “we’re AI native.” One of the things I’ve tried to do in the report is separate the claims. Everyone says they’re AI-enabled, but not all AI is equal. I’ve segregated the report into classifications from zero to four. They’re crude at the moment, but they help separate companies that are no-AI or have just a chatbot feature from real AI rebuilt around models. A sticker is not a strategy — I’m trying to create an objective, vendor-neutral perspective on what really is AI and what isn’t.

19:20Naming the fear: probabilistic AI vs a deterministic profession

Let me talk about the fear, and let’s name it. Fundamentally, there’s a fear that AI will get things wrong. We’ve all seen it hallucinate, forget, run out of context. Generative AI works probabilistically — a bit like a human, guessing as it goes based on what it thinks the answer is. It’s not based on underlying data captured deterministically in a system. And accounting is a deterministic profession. There’s double-entry; everything has to marry up.

So AI in accounting has this tension — you hear it a lot: probabilistic systems won’t work in a deterministic fashion. That can be overused to dismiss AI as hype, because most software providers understand you can’t just make probabilistic guesses about whether a number is accurate. It has to be deterministic. But probabilistic LLMs help get to that answer more quickly and help around the edges. The fear is valid — in an industry where close is not good enough, you’re right to twitch if the numbers are only approximated.

21:40The moat nobody is talking about: liability

The counter side is the moat nobody’s talking about. When we discuss AI’s impact, we usually talk about two moats: vast volumes of data — which Xero and big organisations sit on — and distribution, the companies with existing customer bases to sell into.

But the moat people aren’t talking about is the liability moat. This was the single biggest penny-drop for me developing the report. In accounting, it’s liability. I think it was Greg Sheehan who put it best — a quote Selso Pinto told me about: people don’t use their accountants because they’re silly and can’t do the accounting. They use an accountant to outsource the anxiety of getting things wrong. For me, that summarises everything. The business that wins isn’t the one with the best AI or the vastest dataset or the best distribution — it’s the company that people trust.

24:02Trust takes time to earn

That’s what all this AI hype misses. For AI to replace the accountant, it has to replace trust. Right now, in 2026, who would trust AI software on its own to do your books? But at some point that trust will be earned — once companies have been through audits, once AI has been integrated slowly, once the profession has adapted and roles have changed. That takes time.

It doesn’t matter how fast the models get better. Every bit of research I’ve ever seen in my career at Xero says accountants are the number-one trusted adviser to small businesses — more than lawyers, more than consultants. People trust their accountants with their money. That is the single biggest defensible moat the industry has.

25:39The private equity thesis: small enough to be noble, large enough to matter

There’s a lot of consolidation in the accounting industry. An ageing population of accountants, a lack of people coming in to replace them, fear over AI replacing the core role, a fragmented market, lots of retiring partners. Capital has always been invested in the UK’s accounting industry, but it’s being invested rapidly now.

So who wins? My take: the first firms that apply two ways of thinking. In the back end, the technology is incredible — they use the best tech, with fantastic AI-trained, experienced team members who understand technology at the core. But that technology won’t matter in the front of house. What matters in the front of house is trust — building a brand that takes decades to evolve and grow.

So we won’t just see PE roll-ups creating Frankenstein numbers of different practices stitched together. I think we’ll see communities of like-minded tech businesses come together with amazing technology, consistently applied, not living in fear, with a trusted brand at the front. Small enough to be noble, large enough to matter. That’s the trend I’m seeing in the UK private equity thesis right now.

27:56The Great Reset: fast clocks and slow clocks

When’s it all going to happen? I’ve got a slider in the report — fast clocks and slow clocks. I’ve called this the Great Reset: the time when most people actually feel AI in the accounting profession. Right now the zeitgeist is lots of talk about AI but not much happening in the real world.

This is speculative, and the only thing you can guarantee is that it’ll be wrong. My current prediction is December 2027, with a window between September 2027 and June 2028. The fast-clock variables: orchestration landing — Xero’s just entered, a few others are doing bits, but no one’s owned it; I think that’s months away. Build-cost collapse — I think we’re already past the tipping point; building technology has collapsed by orders of magnitude.

There’s a really important idea: it only takes one. Stuart McLeod said this in the report — it only takes one large firm to do a major reprice, and the whole market suddenly realises it’s in a different ballpark.

The slow clocks have less impact: the accounting bodies and insurers — engaged but not static; and HMRC frequency — they’ve moved to quarterly filing for MTD ITSA and VAT, and at some point they’ll want monthly, then daily, then real-time. And finally, you can’t argue with biology — the composition of accountancy, the humans, is changing. All of that goes into the prediction.

30:57What to do about it, by audience

I’ve got four audiences; I’ll work in reverse.

If you invest in the category: the UK market is mispriced. VCs should back the orchestration cohorts — there’s thin supply in the UK. So much of what people expect is framed as “what software are you on?” — but people didn’t come to get their accounting done by software, they came to get it done by an accountant, by the service. There’s a real rise of tech-enabled service rather than software distinct from service.

If you build software: own a scene, not a screen. Screens commoditise — the sidebar Copilot is an obituary. Owning the scene means owning the infrastructure, the data, the client record, and stitching it together with surface providers, which increasingly look like Claude and ChatGPT. And build for provenance: the audit trail is the new moat. That was a key feature in Xero’s new foray into AI.

If you lead a finance function: stop buying software by features, buy by outcomes. Ask every provider who carries the liability. You’re buying the transfer of anxiety, not a dashboard. The point solution is over — demand a system, not a bolt-on.

If you run a practice: build or adopt the orchestration layer. Try it on one client segment this quarter. And capture the wake code — the intrinsic, unspoken language inside an organisation. Having a digitised version of the decisions that get made is going to be really important, especially if your senior partners plan to retire. Get set up, run trials — nothing better than test and learn.

34:52Exploring the report — and the close

Come and have a look at the report. It’s a live version — not everyone likes dark mode, so there’s a light mode that’s easier to read. You can listen to an AI version of me, bookmark sections, come back anytime. There’s a top-thirty stack, a directory of every provider, and I’m building a recommendation engine and a live news feed so you can stay up to date and save your stack.

I know what it’s like to try to read a twenty-thousand-word report — some people read it and love it, others just ask it anything. “Is the report worth reading?” Let’s hope Anthropic, who support the answer, says yes. And the final question: “What should my firm do from tomorrow morning?” Subscribe to the A to Z of AI Accounting Software — a bit cheeky, but I think it’s a genuinely good resource for navigating the players in the space.

Whether you lead from the front or watch from the sidelines: there’s so much happening here, and around the end of next year there’s going to be a big reset. It’s much more fun to have front-row seats to that than to watch from the sidelines. If you want front-row seats, come and visit the website. And hopefully you got a flavour of what’s possible pulling a website together in Claude.

38:38Q&A: what a practice owner should focus on now

What would you focus on over the next three to six months, and what would you dismiss as hype? First, maybe avoid LinkedIn — though I’m always on it myself. There are extreme perspectives on both sides: “AI is going to replace accountants” versus “it’s all nonsense.” The reality, like anything, is somewhere in the middle. You’ve got to test and learn. You can’t sit on the sidelines. Find some of these companies in the AI space — some of those orchestration engines are really important. Vinyl is a good one too: capturing every conversation is incredibly powerful — you can rip down every bit of conversation and use it for proposal creation, forecasting, or training staff. Try a couple of the best solutions on a few clients. They probably won’t work, but if they do, you’ve unlocked a massive learning. Learn by doing, not by LinkedIn.

40:17Q&A: one tool to start with, and one that’s hype

What one tool would you start using if you weren’t using anything? It depends on your part of the stack, but assuming an accountant looking to automate workflows, I’d say Briefcase. It does some really smart things around capturing and normalising data, rolling it into the general ledger, and coding it. It pulls into HMRC to help with coding, and you can flow information in and post journals. Crucially, you can inform and train it — tell it “that was wrong” — so you’re not letting it loose on your books. A good starting point and a really good bit of kit.

What one tool is hype and won’t help? I’ll be a bit cheeky and answer another way. The most amazing tool everyone should be using is Claude — though OpenAI’s latest model is actually a lot better. What’s hype is using Claude to run your practice. A lot of the tools built specifically for accounting are building deterministic systems with auditability, traceability, and structured data — that’s fundamental. If you vibe code Claude on top of your data without having it built properly, that’s a real concern. I see people saying “I vibe coded an accounting tool on the weekend using Claude,” and there’s real danger in that. You can do meaningful things, but I’d caution people against going down that route — look at the providers laid out in the A to Z and go to conferences like this to learn more. The next session at 10:15 with Dave Selick is literally titled “I vibe code, but you shouldn’t yet” — he’s got a brilliant contrarian view, and there are a few quotes from Dave in the report. We’re trying to get to a pragmatic approach rather than extreme hype on either side.

44:53Q&A: ChatGPT vs Claude, and using multiple LLMs

Should accountants get off ChatGPT and onto Claude? And what about using multiple LLMs? I use multiple LLMs — they all do different things well. Claude, especially Claude Cowork, is a super-structured way of loading up your data and creating skills; an incredible intermediate tool with really solid outputs. I haven’t used Codex, only Claude Code, plus Lovable and Replit and a few others — but I always come back to Claude Code. It wins for me on coding, but it’s slow right now.

Anthropic is the fastest-growing company that’s ever existed — someone said Anthropic grew its core revenue in a month by the total revenue of the three biggest SaaS companies. I’m not sure that’s true and might need to fact-check it, but whatever it is, it’s growing at a phenomenal, exponential pace.

I definitely don’t rely on one model. I use a panel of models for complicated questions. I often use OpenAI to write prompts for Claude. And Google is really good — it has the whole foundation of Google’s algorithms underneath, so it’s great for pulling external research and data on companies. Different horses for different courses — but Claude is where I’m burning most of my tokens.

47:49Q&A: data posture, Copilot, and the limits of expertise

We had a good session earlier comparing Claude versus Copilot for the same tasks — the message is that different models handle things differently and better depending on the use case. Even within Vinyl we use different models for different parts of what we’re doing.

Personally, I’ve tried Copilot and it’s just not for me. But a lot of companies use Microsoft, and keeping your data protected is one of the biggest issues. That posture around data is a really important force — but it conflicts with staying on top of the latest models. Companies have to think hard about how to ring-fence data rather than just defaulting to “Copilot’s the answer,” because that can hold you back massively. That said, I’m no expert — and I think if you say you’re an expert in this, you’re probably a liar, because it’s moving at such a pace. That’s why I’m investing all my time staying on top of these providers; even that’s a full-time job.

49:19Q&A: firms vs software vendors, and why software is dead

How much should firms worry about this versus letting software vendors worry about it? If you’re a software company, your job is to use the right and latest technology, so you have to do it. If you’re an accounting firm, you have a real responsibility for someone in your firm — even if it’s just you — to dedicate a chunk of time to research and to talking to the right software companies that fit your business.

Here’s a punchy statement: software is dead. I believe software as we know it will go away. In accounting, the service — removing the anxiety for your clients, being that adviser and confidant who helps them understand the financial elements of their business — is what people pay for. They don’t pay for clicks and dropdowns and dashboards. They pay for the fundamental service. The firms of the future have to be technology-savvy, just like any other profession, but you can’t leave it to software. You’ve got to invest the time. This is not the same as the cloud wave — I believe it’s a seismic shift, way bigger than anything we’ve experienced in technology. Nothing will replace you, because nothing can replace the trust your clients have in you. Be that trusted adviser at the front, and have that tech-savvy experience in your team if it isn’t you.

52:26Q&A: it’s not just about saving time

There’s a misconception that AI tools are all about saving time and driving margins. But there’s huge value in empowering your team to do better work and delivering a much better client experience. It’s less “I’ll save an hour on every job and make X more money” and more what that means for your team’s work and the experience your clients receive. That’s why we’re all here — to let clients experience things in a better way. That’s what people are willing to pay for, more than any AI credentials. All the AI nonsense will go away; it’ll be about placing trust in someone who genuinely cares about what they do. Better margins help because you can invest more in that — but the fulfilment, and the companies that win this whole wave, come from genuinely caring about the work.