The claim that AI is about to erase administrative and executive-support roles does not match the labour data. US employment in these roles is projected to hold roughly flat through 2034. What is changing is not whether the work exists but where it sits and under what contract: the function is being repriced and redistributed, not eliminated. AI did not replace the judgment in executive support – it removed the coordination overhead that kept that judgment tied to the next desk. The firms that understand this are restructuring support, not cutting it.
There is a widely repeated claim that artificial intelligence is about to erase administrative and executive-support roles. Every quarter produces another forecast of desk jobs automated into obsolescence, another executive survey planning cuts, another essay on the hollowing of the corporate middle.
The labour data does not agree. Not in the way the claim requires.
According to the Bureau of Labor Statistics, employment of secretaries and administrative assistants is projected to show little or no change from 2024 to 2034. The occupation holds roughly 3.5 million jobs, with about 358,300 openings projected each year over the decade. Office and administrative support occupations in aggregate accounted for 18.5 million workers as of the most recent full survey, close to 12 percent of national employment.
That is a stable function, not a disappearing one.
What is changing is not whether the work exists. It is where the work sits, who does it, and under what contract. The function is being repriced and redistributed rather than eliminated, and because headcount stays broadly flat while that happens, the shift barely registers in the indicators most people are watching.
The Number Worth Sitting With
BLS Occupational Requirements Survey data shows that telework applied to only 21.3 percent of secretaries and administrative assistants in 2024.
Roughly four in five administrative support workers in the United States are still physically co-located with the people they support, in 2026, after a global remote-work experiment, after asynchronous collaboration tooling matured, and after AI systems became capable of handling the structured layer of coordination work.
The capability to distribute this function has existed for years. The function has mostly not been distributed. That gap is the story.
What Actually Changed
The constraint was never whether support work could be done remotely. It was whether the judgment-bearing part of it could. Prioritising correctly, reading a relationship, handling a sensitive communication with the right tone, knowing what to escalate and what to absorb – that layer was considered non-transferable because it depended on context and proximity.
AI changed the economics of that transfer. Not by replacing the person doing the judgment, but by collapsing the coordination overhead that made distributed judgment work impractical. Context that once required physical presence and constant synchronous contact can now be carried through documented systems, AI-assisted briefing, and asynchronous protocols. The human still supplies the judgment. They no longer have to supply it from the next desk.
This is the distinction that matters, and it is the one the automation narrative misses. The valuable part of executive support was never the keystrokes. It was the judgment. AI handles the volume around the judgment; it does not replace the judgment itself. A virtual executive assistant operating with modern tooling delivers more capability per hour than an unassisted equivalent did five years ago, which is precisely why the function is becoming portable rather than disposable.
The Evidence on What AI Does to the Seat
The most rigorous field study available looked at 5,172 customer-support agents at a Fortune 500 firm, published in the Quarterly Journal of Economics. Access to a generative AI assistant raised productivity by roughly 15 percent on average, measured as issues resolved per hour.
The average is the least interesting figure in the paper. Gains ran to 34 percent for novice and lower-skilled workers and to approximately zero for the most experienced and highest-skilled, who saw small speed improvements alongside small quality declines. The study also found that an agent with two months of tenure using the tool performed at the level of an agent with six months of tenure without it, and that AI assistance measurably improved communication quality, with the effect concentrated among less-experienced workers.
The implication for anyone building a support function is direct.
AI raises the floor faster than it raises the ceiling.
It brings a capable but less-tenured operator up the experience curve quickly, which means a well-structured, well-supported team reaches high output faster than the old training timeline assumed.
Why This Is an Opportunity, Not a Threat
For most small and mid-sized firms, the takeaway is not that their people are about to be automated away. It is that the fixed, local, full-time structure of executive support is no longer the only option, and often no longer the best one.
The historical choice was a full-time in-house hire carrying salary, employment taxes, and management overhead, justified only if the workload stayed consistent all year. It rarely does. The alternative now available is a retained arrangement that delivers senior support capability without the fixed payroll line, scaling up through intensive periods and back down when things quieten.
At the smaller end of the market this is already the dominant pattern. Boutique investment firms, family offices, independent advisory practices, and founder-led businesses have been quietly restructuring their support functions toward distributed executive support for several years. The model converts a fixed cost into a variable one and gives access to a calibre of support that was previously the preserve of larger organisations with the budget for senior in-house staff.
What It Means for the Next Decade
Three consequences follow.
The operating leverage of small firms improves. The fixed cost base that historically made sub-scale firms fragile in a downturn becomes partly variable, which changes the survival maths across a long tail of small businesses.
The demand for local, co-located administrative headcount erodes slowly and quietly rather than through visible layoffs, because the shift shows up as roles never created rather than roles cut.
And the policy conversation, calibrated around a headcount-destruction model, is aimed at the wrong phenomenon. The workers affected are not being replaced by software. They are being competed with by a distributed labour pool that software made accessible, and supported by tools that close the quality gap from the bottom.
The narrative says AI is coming for these jobs. What is actually happening is that AI removed the last practical constraint on distributing them, and the market is now working through a repricing that will take a decade and barely register in the statistics everyone is watching.
The firms that understand this early are not the ones cutting support. They are the ones restructuring it.
