Victor Avendaño
EspañolContact

AI & Productivity ·

Will AI Replace Your VA or Your PM? I Read the Research So You Don't Have To

If you’re about to hire a virtual assistant or a project manager, I’d bet you’ve already had this thought: can’t AI just do that now?

Fair question. And I didn’t want to answer it with my opinion, because of course I’d say no. I do this for a living. So I went and read. Ten reports and studies, from economists, consulting firms, a big tech survey, and a freelance platform with its own contract data.

Here’s what I took from them, in plain words.

The short version: AI is taking over tasks. It isn’t taking over responsibility. The catch is that a lot of those tasks are exactly what many assistants and PMs spend their week on.

What is actually going away

Let’s start with the scary headline, because it’s real. The World Economic Forum’s Future of Jobs Report 2025 expects administrative assistants and executive secretaries to be among the roles that shrink the most, and it lists AI and information-processing technology as one of the reasons.

Now the part that doesn’t make the headline. In that same report, project managers are among the job categories adding the most jobs.

Same report, two arrows pointing in opposite directions. Why?

Gartner gave a hint back in 2019. They predicted that by 2030 AI would take over about 80% of today’s project management tasks, and the examples they gave were data collection, tracking and reporting. Notice what’s missing from that list. Nobody said AI would decide which risk is worth taking, or sit across the table from a difficult stakeholder. It’s a 2019 forecast, so I read it as a direction, not a date. But the direction matches everything else I found.

Harvard Business Review made a similar case in 2023. The authors open with a number I keep coming back to: only about 35% of projects are completed successfully. Their argument is that the technology project teams rely on is still immature, and AI is about to change that. The way I read it: better tools for the tracking and record-keeping, so people can spend their time on what decides whether a project actually lands.

So the pattern isn’t “assistants lose, PMs win.” It’s that the collecting, copying and reformatting part of any job goes first. If your job was mostly that, you have a problem. If your job was mostly deciding what to do with it, you don’t.

It’s not about job titles

Upwork looked at its own platform data and found something that surprised me. Freelancers in administrative support saw their earnings go up 6 to 7%, which Upwork attributes to generative AI. Writing and translation went down, 8% and 10%. Customer service went down too, 4%.

Same technology, opposite results. Upwork’s explanation is that AI makes routine, low-value work cheap and makes the higher-value work that sits on top of it more valuable.

A small warning: that’s one platform and mostly one year (2023), and it’s a different kind of evidence than the WEF’s survey of employers. Neither is the last word. But underneath, they say the same thing. The question isn’t “which job,” it’s “which tasks, and how easy is it for a machine to do them.”

What happens to the work that stays

Two studies I’d actually recommend reading.

In the first, researchers followed more than 5,000 customer support agents who got an AI assistant. On average they resolved about 15% more issues per hour. The people who gained the most were the least experienced. The most experienced agents barely sped up and, if anything, their quality dipped a little.

In the second, a randomized experiment with college-educated professionals doing realistic writing tasks, ChatGPT cut the time by about 40% and raised quality ratings by about 18%. Something else I liked: people spent less time on rough drafts and more time coming up with ideas and editing.

Put those together and here’s how I read them. AI lifts the floor. “Can write a decent email” or “can summarize a call” stops being something you can charge for. What’s left is the part the tool can’t do for you: knowing what to keep, what to cut, and what to question.

Where it quietly goes wrong

This is the one I’d keep in mind. A team of researchers working with BCG gave 758 consultants a set of realistic tasks, some with GPT-4 and some without. On tasks the AI handles well, consultants finished about 12% more of them, about 25% faster, with quality rated around 40% higher. Impressive.

But the team also built one task where the AI would give a convincing, wrong answer. People working without AI got it right 84% of the time. People using AI got it right 60 to 70% of the time.

The researchers call the edge between “AI is great at this” and “AI is confidently wrong at this” the jagged frontier. The problem is you can’t see where it is. And the tool sounds so sure that people relax and stop checking.

Then there’s the 2024 Work Trend Index from Microsoft and LinkedIn. 75% of knowledge workers already use AI at work, and 78% of those bring their own tools. 60% of leaders say their company has no plan for it. Only 39% of users have received any AI training from their employer.

So the risk isn’t that your assistant uses AI. They probably do already. The risk is that nobody agreed on how.

If I were the one hiring

This is what I’d ask. And it’s what I’d expect to be asked.

  • “Where do you use AI, and where don’t you?” A good answer names both. “Everywhere” and “never” are both red flags.
  • “How do you check it?” Who verifies what, and what happens when the tool is confidently wrong.
  • “Is your process written down?” If the work lives in one person’s head, AI makes it faster, not safer.
  • “What do you own?” Outcomes, not hours of drafting. A clean inbox. A plan that holds. A report you can trust.

What I do with all this

I use AI for first passes: drafts, research summaries, spotting patterns in a pile of cases. Then I own the final version, and I can tell you what I checked.

At ARQ, I designed an AI-based first-pass filter for source-of-funds verification. It cut unnecessary manual review by over 90% and gave users faster feedback on missing documentation. The word to notice in “first-pass” is first. The point was to clear the obvious, so people could spend their time on the cases that needed judgment.

If you want the day-to-day side of this from the project manager’s chair, I wrote about it in How AI Is Changing the Project Manager Role.

If you’d like to look at your own week and figure out which parts are first-pass work and which parts need a person, message me. We can go through it together.

What I read

Ten sources, eight of them quoted above. Figures are from the versions linked here, and a few of these started as working papers that were revised later.

The two I didn’t quote: McKinsey’s 2023 report on the economic potential of generative AI, and ASAP’s 2025 survey of administrative professionals. I couldn’t open the first one’s original, and the second keeps its numbers behind a membership.

← Back to blog