Which virtual assistant work is actually exposed to AI

This article was contributed by Alon Pearl, Founder and CEO of VA Masters. He founded the firm in 2021 after hiring a virtual assistant remotely to solve his own bandwidth problem, and now writes on hiring at volume, drawing on the 166,350 applications the firm processed between January 2025 and July 2026.

An honest answer to the hardest question in this category, and what the job descriptions show.

The question deserves a straight answer

Anyone running a business that depends on remote support work has asked some version of this, and most of the answers available are either dismissive or alarming.

I would rather give the one I actually hold, including the part that is uncomfortable.

“Tasks which are repetitive and clearly defined are the most exposed to automation. But we’re also seeing the opposite happen. New roles are emerging around AI itself, whether that’s administrative people who know how to work well with AI tools, AI-assisted content creation, image and video editing, or analysis work. In almost every role we recruit for today, employers want someone who knows how to use AI.”

The first sentence is the one that matters. If someone tells you nothing in this category is exposed, they are selling you something.

What the job descriptions show

We looked at 322 roles opened between February and August 2026 to see where employers are writing AI into the requirements. 23.9% name it in some form.

322 roles opened February to August 2026. Selected categories; those with fewer than 15 roles are not reported.

One thing has to travel with that table and it is not a formality.

We are describing where employers ask for AI skills, not where AI is capable of doing the work. Those are different questions and this data answers only the first.

With that said, the pattern is the opposite of what the public conversation assumes.

The categories asking for AI most often are the ones that already require judgment. The categories built on defined procedure sit at the bottom, with finance and accounting lowest of all at 5.3%.

Administrative and operations support, the largest category by volume and the one people picture first when they imagine AI taking over office work, sits below the overall average.

Why ambiguity is the real line

“An AI tool onboards in minutes; a human VA takes days to weeks. But that comparison is misleading. AI wins on speed and simple, repeatable tasks. The moment the work needs judgment, context, or catching an edge case the AI misses, a human pulls ahead fast. In our experience the tipping point isn’t task volume, it’s task ambiguity. The messier the decision, the more you need a human.”

That is the test I would apply to any role you are thinking about. Not how much of it there is, but how much of it is decided in advance.

It also explains something we found elsewhere.

In our analysis of the software these roles name, roughly a quarter of roles named no software at all.

Those are usually roles where the process has not been built yet, which is precisely the kind of work that does not automate, because there is nothing defined to automate.

What is growing rather than shrinking

“It’s a bit like driving. Knowing how to hold the wheel properly with these tools gets a client far better, more productive results. If anything, AI is making virtual assistants more accessible, not less. It helps bridge gaps that used to be real barriers, both language gaps and skills gaps.”

“Someone who only did repetitive copy-paste data entry is exposed. But someone doing customer service today, who adapts and understands how to manage the tools that cut down on simple repeat inquiries, becomes more technical and starts suggesting automations themselves.”

What employers increasingly look for is people who know how to learn, on the assumption that someone with a sharp head and an AI tool will go and solve the problem themselves.

That is a different hiring criterion from the one most job descriptions are still written around.

What this means if you are hiring

Describe the ambiguity, not just the tasks. If the role involves decisions that are not settled in advance, say so in the posting. That is the part of the job a tool will not take, and it is what you are actually paying for.

Ask for the skill, not the tool. In our data only 10.2% of roles name a specific AI product. Most employers asking for AI are asking for general capability, which is a far easier requirement to hire against than a particular piece of software.

Do not write an AI requirement you cannot explain. A line asking for AI proficiency in general tells a candidate nothing and filters nobody. A requirement that names a task, such as drafting first versions or cleaning data before it reaches a report, tells them something.

Treat the ability to learn as a qualification. It is already how a growing share of these roles are being filled, whether or not the job description says so.

About this data

The figures here describe what passes through a single recruitment pipeline, not the offshore hiring market as a whole.

Every dataset of this kind reflects the business that produced it. Our client base skews toward the companies that find us, most of whom arrive on their own through search, through AI assistants, and through referrals from existing clients.

One bias is worth marking explicitly.

Clients who reach us through search and through AI assistants are likely to be more comfortable with technology than the average small business, which may lift these figures above what a random sample would produce. We cannot quantify it from our own data.

Workflow automation platforms are counted separately from AI throughout and are not included in any AI figure. Roles opened by the agency for its own team are excluded.

Every figure carries its own window. Full methodology sits alongside our Filipino virtual assistant workforce dataset, and the shape of the roles themselves is covered in our analysis of what businesses actually outsource.