Right now almost every software vendor in accounting is claiming AI leadership, and shipping something that looks impressive has never been easier. But accountants do work that businesses can’t afford to get wrong. Speed is the wrong thing to compete on when the real test is whether an accountant can take professional responsibility for what the software produces. In accounting someone signs off on the final work and answers for it if it’s ever questioned, and that responsibility doesn’t disappear because AI was involved.
The hard part is invisible
Building an AI feature that demos well in a controlled setting is the easy part. The hard part is everything underneath: the testing, the guardrails, the edge cases, the controls that stop the system doing something wrong the moment it meets real client data. None of that shows up in a polished demonstration and you don’t find the limits of an AI product until you’ve had your hands on it in conditions nobody anticipated.
There’s a related discipline that gets skipped. We reach for AI when it’s genuinely the best way to solve a problem. Plenty of things are handled more reliably by well-established methods, and using a model where a simpler approach would do just piles on risk, training and governance overhead for no real gain. The point is to fit the tool to the job.
The bar is work you can stand behind
Everyone can call a frontier model. Not many can hand an accountant something they’ll put their name under. That’s the bar we hold ourselves to, because compliance work is judged on how you got there as much as the answer itself. It needs the trail that makes the answer defensible, and a generic AI product has no way to produce that.
Clear that bar and something more useful becomes possible. There’s never enough time to do the work for every client, to the depth every client deserves, so the deepest attention has always been rationed to the top tier. If an agent prepares the work and the accountant can review and approve it faster than doing it from scratch, that depth stops being a luxury. That’s what we’re building towards, and it’s why a fast product that an accountant couldn’t stand behind would miss the point entirely.
Not rushing doesn’t mean standing still
Being deliberate about what we ship hasn’t meant sitting on our hands. We’ve had dedicated research time on AI throughout.
The way we like to innovate is through prototypes: build something quickly, put it in front of a small number of customers, listen hard, and let that steer what comes next. That’s already happening with our latest innovation in agentic AI. We’re keeping those pilots close while we learn, because the goal is to arrive at something that’s a real benefit to the people using it.
For firm leaders weighing this up
Scepticism about AI is reasonable. There’s been genuine hype ahead of what the technology can currently do and the claims deserve scrutiny. But scepticism shouldn’t harden into inertia, because the real effects are already here and the gap between the hype and the reality is closing faster than most expected.
For accountants, whose expertise is deep and genuinely valuable, it picks up the work around the edges of that expertise so you can do more with the same time. Professional judgement stays where it is. Accounting firms are well placed for this, having already navigated the move to cloud and constant regulatory change. This is one more change to adapt to, and a less dramatic one than it’s often made out to be.
Where we go from here
We’re not rushing, and that’s deliberate. The aim is straightforward: the accountant doing the deep work for every client, not just the top tier, with an agent preparing it and the accountant still the one who checks it and takes responsibility for it. The accountant stays the hero. Our job is to have their back, and to be deliberate enough to earn that. In the next post we’ll take you inside the pilots where we’re testing whether we have.
Follow the series on our agentic AI hub, where we go inside the pilots we’re running and what we’re learning.
