The short answer

Some of it, yes. Most of it, no. And not in the way the question implies.

The work that goes is a specific kind: repetitive, high-volume, low-judgement. Typing what is printed on a document. Matching the obvious bank lines. Writing the same email again.

The work that stays is most of what people were actually hired for.

Here is the detail, because the general answer is not much use to anyone deciding what to do.

What is genuinely going

Six tasks, named specifically.

Typing supplier invoices. The system reads them. A person checks the draft.

Matching obvious bank transactions. Two hundred clear matches proposed; a person handles the twelve that are not clear.

Reading long threads before a call. Summarised.

Writing routine follow-up emails. Drafted, then edited.

Producing repetitive documents. Descriptions, standard quotes, support articles.

Assembling reports for a meeting. The gathering, not the discussing.

Notice what these have in common. Every one is work people describe as necessary rather than valuable. Nobody’s professional identity rests on typing invoice lines.

What is not going

Five categories, and they are larger than the first list.

Checking

Every AI output in a business system needs review. Extraction is good, not perfect. Predictions are estimates. Generated text is fluent, not necessarily true.

This is new work, not old work removed. Somebody has to do it, and it needs the same knowledge as doing the task did.

Exceptions

The twelve unmatched bank items. The invoice that does not match the receipt. The customer whose situation fits no pattern.

Exceptions are where the difficulty always was. Automation removes the easy 90% and leaves the hard 10%, which was already the part that needed a person.

Judgement

Whether to extend credit. Whether a provision is adequate. Whether a supplier’s explanation is credible. Whether this deal is real.

Relationships

Difficult conversations. Negotiations. A customer who is upset. A supplier who needs managing.

Accountability

Somebody signs the accounts. Somebody answers to the board. Somebody explains to the customer.

That person has to be able to explain the numbers. “The system did it” is not an answer, and knowing that shapes how the work has to be done.

Why “replacement” is the wrong frame

The question assumes jobs are bundles of tasks that either survive or do not.

Jobs are bundles of tasks with different weights. An accounts role might be 40% data entry, 30% checking and investigation, 20% reporting, 10% advising.

Automate most of the first, and the role does not disappear — its composition changes. Less entry, more investigation, more advising.

For most people that is a better job, and it is closer to what they trained for.

The genuine exception is a role that was almost entirely data entry. That changes substantially, and it is worth being straight about. In practice, retraining into checking and exception work is usually both possible and cheaper than replacement.

What actually happens in small and medium businesses

Worth separating from the enterprise version of this conversation.

Most SMEs are not overstaffed on admin. They are behind on it.

Invoices posted late. Customers not chased. Suppliers not reviewed. Reports not produced because nobody had time.

When capacity is released, it usually goes into work that was not getting done rather than into a reduction in headcount. That is not a comforting story — it is what tends to happen at that scale.

The realistic outcome for a thirty-person company: several roles get less tedious, some overdue work starts getting done, and nobody leaves.

What this means for your team

Five things worth doing.

Be specific. “AI will make us more efficient” sounds like “we need fewer of you”. “You will spend less time typing invoices and more time investigating variances” does not.

Say which case applies. If a role changes substantially, say so early and plan the retraining. People fill silence with the worst version.

Show the errors. Do not present it as flawless. It builds appropriate trust and it makes clear why the checking role matters.

Say the review is permanent. If people believe review is a formality before removal, they will stop doing it properly.

Name what the freed time is for. Otherwise released capacity quietly fills with more of the same.

The uncomfortable part

Two honest points.

Some entry-level roles are genuinely affected. The job that used to be “type these invoices while you learn the business” is thinner than it was. That was often how people entered a function, and the path in is narrower.

The checking role needs more knowledge, not less. Reviewing an extracted invoice requires knowing what is correct. That is a higher bar than typing what is printed, and it changes who can do the job.

Neither of these is a reason not to adopt. They are reasons to think about how people learn the work.

The longer view

Will this change further? Yes. The direction is toward systems that prepare whole pieces of work rather than single tasks.

But the commit step is likely to stay with people for a while, because reversal in a business system is expensive and someone has to be accountable for the numbers.

Predicting beyond that is speculation. What is safe to say: the work that survives is the work involving judgement, exceptions, relationships and accountability — and that is a description of most professional work, not a small residue.

The short version

Yes, some manual ERP work is going. Typing, matching, assembling, drafting.

No, jobs are not being replaced in most small and medium businesses. Roles are changing composition, and the parts that go are the parts nobody valued.

The new work is checking, and it needs more knowledge than the work it replaced.

Be specific with your team about which tasks change, keep the review permanent, and name what the freed time is for.