The data point that sits underneath everything written about AI and administrative assistants in 2026 is the US Bureau of Labor Statistics projection: a 7% decline in the broader category of secretaries and administrative assistants from 2024 to 2034. That is one of the steepest negative projections in any white-collar occupation. It is not balanced by growth in another sub-category. The role, in its general form, is contracting.
What that headline misses is that the contraction is not happening evenly. The general office assistant role is in steep retreat. The executive assistant supporting senior leadership is showing real resilience. The story of the profession in 2026 is the bifurcation between those two, and where on the spectrum a working assistant sits.
The work AI is genuinely automating
The tier-1 administrative work that has historically filled an admin’s day is the most fully automatable category of office work. Data entry, calendar coordination through scheduling links, simple document formatting, basic email triage on routine inboxes, expense report submission, and travel booking on standard itineraries are all tasks where AI tools (often inside Microsoft 365 or Google Workspace by default) close the loop without the human touch they used to require.
The Anthropic Economic Index names office and administrative tasks as one of the highest-exposure categories on its data. Drafting and editing, summarisation, document conversion, and information lookup, all of which used to be a meaningful share of an admin’s week, are exactly the categories where Claude usage is highest.
This is the category of work that is shrinking. Not slowly. Headcount in routine clerical and general office support has been falling at most large enterprises, generally through attrition and non-replacement rather than active layoffs, but the trajectory is unmistakable in BLS data and in the public hiring data from companies like Indeed and LinkedIn. The job posts that used to specify “general administrative support” are increasingly absorbed into other roles or replaced with self-service systems.
The work AI is not automating
The executive-support layer of the profession looks different in the data and on the ground. The work involves judgement about people, context that cannot be put into a prompt, and trust relationships that take years to build. The same AI tools that automate tier-1 work are being used by executive assistants as a force multiplier, expanding what one person can support rather than replacing them.
What an executive assistant supporting a CEO actually does on a Tuesday includes things AI cannot do. Knowing which of three competing meeting requests to push back on, given a week the principal does not yet know is going to go sideways. Handling a difficult call from a board member with the right tone, judgement, and discretion. Briefing the principal before a hard conversation in a way that respects what the principal already knows and does not need explained. Understanding which travel route will leave a chronically jet-lagged executive sharp on landing.
These are not edge cases. They are the daily core of executive support. The data and the working pattern both confirm that this part of the job is holding up, and in some cases growing in scope. Many EAs in 2026 are taking on chief-of-staff-adjacent work, owning project management, communications, and decision tracking that used to be split across multiple roles.
Why the bifurcation is widening
Two forces are pulling the two ends of the profession apart at the same time.
On the tier-1 end, the cost of automation keeps falling. A scheduling agent that costs USD 20 a month per user is now equivalent to a meaningful share of a part-time admin’s time. The economics of automating routine clerical work at scale have moved past the tipping point at most large organisations. The displacement is structural, not cyclical.
On the executive-support end, the value of a great EA has gone up rather than down. When the routine work is automated, the principal is dealing with more inbound, more decisions, and more demands on their time, not fewer. Inboxes are noisier because every counterparty also has AI to draft more email. Calendar requests pile up faster because everyone is more efficient at proposing meetings. The screening, triage, and trust-based filtering an EA does is more valuable when the volume is up and the signal-to-noise is down.
The result is a profession where the bottom is hollowing out and the top is consolidating. The middle, which used to be the bulk of the role, is the part most exposed.
What the working admin should actually do
For an admin currently in a tier-1 or generalist role, the pragmatic moves cluster on three tracks.
Build AI fluency on the tasks the role still includes. Get comfortable with using ChatGPT, Microsoft Copilot, or Claude for drafting, summarising, and document work. The point is not to do less. It is to do more in the same time, so the role’s scope can grow rather than shrink.
Move toward executive support deliberately. The roles that are durable are the ones supporting senior leadership in a relationship-heavy way. The pathway from generalist admin to executive assistant is real, and the people who walk it are not displaced. The pathway runs through demonstrated judgement, written and verbal craft, and the ability to be trusted with sensitive context.
Take on adjacent scope. Project coordination, internal communications, light financial admin, and event management are all categories where an admin who has freed up routine time can credibly absorb work the organisation needs done. The admins who treat AI as their assistant rather than their replacement are the ones building larger remits.
What the next twelve months change
The structural picture does not change quickly. BLS does not change its projection on a one-year cycle. What does change is the local picture inside individual organisations.
Mid-2026 to mid-2027 is likely to be the period when the larger enterprises finish the first round of post-AI org redesigns. The function that has historically housed centralised admin pools (often under HR or operations) will look different by the end of 2027 in most large companies. The admins who have invested in AI fluency and in moving up the value stack will be in a different conversation than those who have not.
The job is not going away. The middle is shrinking and the top is growing, and the difference between being on the right side of that line and the wrong side comes down to deliberate skill-building over the next year. The hard part of the AI shift for this profession is not the technology. It is the choice each working admin has to make about what their next role is going to look like.