Survey data on how accountants feel about AI keeps finding the same thing, and it is not what most partners expect.
Ask a managing partner why the firm’s AI rollout has stalled and you will usually get some version of the same answer. The technology is there. The licences are bought. The problem is people, and specifically the people who will not get on board.
It is a tidy explanation and the evidence does not support it. What the survey data actually shows is a firm in which almost everyone is more positive than their colleagues believe, a group of staff who are considerably more anxious than their partners register, and a conversation that nobody starts because each side has misread the room.
A caveat belongs here first, and the presenter of the masterclass this draws on makes it himself before quoting a single number. This is vendor research, published by a practice management software company with a commercial interest in firms adopting AI, and it should be read as directional rather than settled. That is an unusually honest framing and it is the right one. What follows treats the findings as evidence worth weighing, not as proof.
The First Assumption: Our People Are Resistant
They are not, or at least far less than the framing suggests.
Across roles, the research found roughly seven in ten professionals expecting AI to bring substantial change to the profession, and among individual contributors, the people actually doing the bookkeeping, payroll and tax work, around 81 percent described themselves as intrigued or excited in the more recent wave. Skepticism among that group fell to about 11 percent year on year.
That is not a workforce digging in. That is a workforce waiting to be given something to do.
Figure 1. Vendor research presented in the masterclass and Karbon’s 2026 report, which covers roughly 600 professionals across six continents. Directional, not definitive.
The 2026 edition, covering roughly 600 professionals across six continents, reports 98 percent of firms now using AI in some form, with most using it daily or several times a day. Whatever the barrier is, it is no longer willingness.
The Second Assumption: Everyone Is On Board
Also wrong, in the other direction, and this is the finding worth sitting with.
Enthusiasm is not evenly distributed, and it tracks almost perfectly with how insulated your role is from the consequences. Among firm owners and leaders, 89 percent landed in the top two boxes of a five‑point scale on excitement about AI, with 8 percent in the bottom two. Among individual contributors, that bottom‑two figure was 18 percent, more than double.
The starker split is on job security. Before you look, make a guess.
Nearly a third of the people doing the work think the work may not exist. One in eight partners agrees. Those two groups are in the same meetings, nodding at the same slides, and holding materially different views about what is being decided.
The Third Assumption: I Am the Only One Excited
This is the finding that explains the other two, and it is a textbook social phenomenon rather than an accounting one.
Asked how they feel about AI, most professionals report enthusiasm. Asked how they think their colleagues feel, they assume considerably less. Almost everyone is positive; almost nobody believes the person next to them is.
The consequence is predictable and expensive. If you believe you are the lone enthusiast in a skeptical firm, you do not raise the idea in the partner meeting. You do not volunteer to pilot the tool. You certainly do not admit you have been using it already. Everyone waits for permission that everyone else is also waiting to give.
If you believe you are the lone enthusiast in a skeptical firm, you do not raise the idea. Everyone waits for permission that everyone else is also waiting to give.
There is independent evidence that the private enthusiasm is real. Work by Erik Brynjolfsson and Avinash Collis at Stanford measured what people would need to be paid to give up generative AI tools for a month, and arrived at a consumer surplus of roughly $97 billion in 2024, against about $7 billion of actual US revenue booked by the major providers. The value is overwhelmingly landing with users rather than vendors, which is a reasonable proxy for how much individual professionals quietly rely on these tools already.
The Fourth Assumption: This Is a Technology Decision
| What the firm is buying | What the evidence says decides it |
|---|---|
| Licences and seats | Training, in cohorts, with hands on keyboards |
| A tool selection process | A written policy that permits rather than restricts |
| An announcement at the partner meeting | Leaders visibly using it themselves and failing in public |
| A pilot in one department | A champion in every department, comparing notes |
| A rollout plan | A conversation about who thinks their job is at risk |
Figure 3. The right column is not more expensive than the left. It is harder, because it requires partners to spend attention rather than budget, which is the scarcer resource in most firms.
The recurring theme across all three report editions is that interest runs well ahead of capability, and the gap is training. Firms investing in structured training, documented strategy and explicit governance consistently report stronger outcomes: more time saved, higher confidence, better results. Firms that bought licences and waited report the opposite.
Two secondary findings sharpen the point. Concern about data security has risen to 83 percent, up seven points year on year, which is what you would expect as usage moves from experimentation to real client data and is a reason for policy rather than hesitation. And 91 percent say graduates are more likely to join a firm that uses AI, which turns this from an efficiency question into a recruitment one in a profession that has had a talent shortage for two decades.
What to Actually Do About It
The practical implication is smaller than a transformation programme and more uncomfortable than one.
Ask your people what they think, in a format where the answer can be honest. Not whether they are excited, which everyone will say. Whether they think the work they do will exist in five years, and what they would want to be doing instead. A third of them may be carrying an answer they have never been asked for.
The technology is not the constraint and has not been for some time. The constraint is that two groups of people in the same firm are holding different conversations about the same future, and neither has been invited to the other’s.
To explore the research in full, including how perceptions have shifted across three annual reports and what the firms getting results are doing differently, watch The State of AI in Accounting on Miles Masterclass.