Will AI replace accountants?

What AI already does well in accounting

Start with the honest part, because the profession has a habit of dismissing this and then being surprised. The mechanical layer of accounting work is genuinely being automated, and not in a demo. Extraction tools read invoices, receipts and statements and produce structured data. Coding suggestions get better the more history they see. Drafting assistants turn a situation and a few facts into a client letter in seconds. Long documents get summarised in the time it takes to make a coffee.

None of that is intelligence in any meaningful sense. It is pattern work at speed, and pattern work is a large share of the hours a practice bills. That is why the question keeps coming up, and why answering it with 'AI can't really think' misses what is actually happening to the timesheet.

• Extracting figures from documents and pushing them into a ledger or workpaper.
• Suggesting transaction coding based on what your firm has done before.
• Flagging anomalies in a dataset far faster than a person scanning a report.
• Drafting the first version of almost any written client communication.
• Summarising contracts, deeds, correspondence and management reports.
• Turning meeting notes into structured file notes with actions attached.

What it cannot do, and will not soon

Every task that survives has the same shape: someone has to be accountable for it. AI has no professional standing, no insurance, no registration and no capacity to be sanctioned. You cannot put a model in front of a regulator, a lender or an unhappy client. That is not a technical limitation waiting for the next model release, it is a structural one, and it is where the value of the profession has quietly relocated.

The other thing AI cannot do is know what your client has not told you. A large part of an accountant's value is noticing the thing that is missing from the file, and asking a question the client did not want asked. Models work with what is in front of them. Practitioners work with what should have been.

What the next few years plausibly change

Three shifts look likely, and none of them are 'accountants disappear'. First, the compliance-hours business gets cheaper to deliver, which puts pressure on any firm whose pricing quietly assumes those hours. Second, the entry-level pathway changes: a lot of the work graduates learned on is exactly the work being automated, so firms have to teach judgment more deliberately than they used to, because it will no longer be absorbed by osmosis over two years of data entry.

Third, and most usefully, advisory capacity opens up. The advisory work most practices say they want to do more of has always been squeezed by compliance deadlines. If compliance takes fewer hours, that constraint loosens, and the firms that have built the skills to use the space will take it.

On the specific version of this question people search for: will CPAs be replaced by AI by 2030? No. Registration, professional standards and legal liability do not evaporate in four years, and the regulatory system has no appetite for accepting a model as the responsible party. What will change by 2030 is the mix of work inside the role, and how many people a given volume of compliance requires.

What accountants should do now

The useful response is not a strategy offsite. It is a handful of concrete moves, in order, small enough that you can start this month.

1. Use the tools yourself. Not a demo, a fortnight of real work. Opinions about AI from people who have not used it seriously are worth nothing, and that includes confident ones.
2. Pick the mechanical process that annoys you most and measure it: how many hours, how many people, how many corrections. You cannot tell whether a tool helped without a before.
3. Write the firm's data rules down. One page. Which tools are sanctioned, what can be pasted where, who to ask. Staff are already using these tools; the only question is whether they are doing it on your terms.
4. Teach judgment earlier. If juniors do less rekeying, they need review work, exceptions and client contact sooner than your old training plan assumed.
5. Look at your pricing. If your fees are built on hours that are about to halve, that conversation is coming whether you start it or not.

The honest risk

The version of this question that deserves attention is not whether AI replaces the profession. It is whether your practice ends up on the wrong side of a widening gap. Two firms with the same clients and the same staff will, within a few years, have visibly different capacity, and the difference will be whether they did the boring adoption work early or waited for it to be settled.

None of that requires a large budget or a technical background. It requires deciding what to use, in what order, and what to ignore. That is the whole job I do, and the reason I do it as advice rather than as a build: the wrong first project is a much more expensive mistake than the software.