In 2013, two Oxford researchers published a number that scared an entire profession: A high probability that computers would automate the executive assistant role within two decades. More than a decade later, the job hasn’t disappeared. It’s changed shape, the same way bank teller jobs changed shape when ATMs arrived and didn’t actually shrink the profession the way everyone predicted. This article walks through what the research on automation actually says, what’s already changing in this specific role, and what it means for how you hire one today.
Key Takeaways for Business Leaders
- A famous 2013 Oxford study rated executive assistant work among the most automatable occupations in its entire dataset, a finding that’s been widely misread as a prediction the job would vanish.
- Economic research on ATMs found bank teller employment didn’t shrink after automation arrived. Tellers shifted from cash handling to relationship and sales tasks, a pattern directly relevant to what’s happening in executive support today.
- MIT economists Daron Acemoglu and Pascual Restrepo distinguish task displacement from task reinstatement: automation removes some tasks from a job while creating new ones, rather than eliminating the role outright.
- AI tools today reliably handle scheduling suggestions, drafting, and summarization- the routine, pattern-based slice of the job- while judgment, discretion, and representing an executive remain entirely human tasks.
- The World Economic Forum’s ongoing research on the future of jobs consistently frames the shift as human-machine collaboration, not human replacement, for roles built around judgment.
- Hiring today means screening for exactly the skills that don’t automate, plus comfort directing AI tools rather than competing with them, a combination the job description itself needs to reflect.
Will AI Replace Executive Assistants? What the Research Actually Says
The Oxford Study That Started the Panic: Frey and Osborne’s Automation Probability Model
Economist Carl Benedikt Frey and machine learning researcher Michael Osborne, both at Oxford’s Martin School, published a study in 2013 titled “The Future of Employment: How Susceptible Are Jobs to Computerisation?” analyzing 702 detailed occupations and assigning each a probability of automation based on the tasks involved. Administrative and secretarial occupations scored among the highest in their entire dataset, with probability estimates above 0.8, placing this category alongside telemarketing and data entry as among the most computerizable jobs they studied.
Why a High Automation Score Doesn’t Mean a Job Disappears
That number measures something narrower than most people assume: the technical feasibility of automating specific tasks within a job, not a prediction that the job itself disappears. Frey and Osborne’s model looked at task composition, how much of the work is routine and codifiable, not at whether an organization would actually choose to eliminate the human once some of those tasks got automated. More than a decade on, executive assistant employment hasn’t collapsed the way a naive reading of that number would suggest, and the reason why has a well-documented precedent in a completely different industry.
What Happens When a Task Gets Automated? Lessons from the ATM
James Bessen’s Research on ATMs and Bank Tellers
Boston University economist James Bessen studied one of the most direct historical parallels available: the introduction of automated teller machines starting in the 1970s. Conventional wisdom predicted ATMs would gut bank teller employment, since the machine automated the single most time-consuming task tellers performed: cash withdrawals and deposits. Bessen’s research found the opposite happened. Because ATMs lowered the cost of running a bank branch, banks opened more branches, and total teller employment kept growing for decades after ATMs became standard. What changed wasn’t the number of tellers. It was what tellers spent their time doing, shifting from manual cash handling toward customer service, account troubleshooting, and selling additional banking products.
Applying the Same Logic to Executive Assistants and AI Scheduling Tools
The same mechanism is playing out with executive assistants and AI tools today. AI absorbs the scheduling suggestions, the meeting summaries, the first draft of a routine email, the mechanical slice of the role Frey and Osborne’s model correctly identified as automatable. What it hasn’t absorbed is the judgment call about which meeting actually matters, the discretion around a sensitive message, or the relationship management that made a great assistant valuable in the first place- the exact tasks Bessen’s research would predict expand to fill the time the automated tasks used to occupy.
Displacement vs Reinstatement: Why Automation Creates New Tasks, Not Just Fewer Jobs
Acemoglu and Restrepo’s Task-Based Model of Automation
MIT economist Daron Acemoglu and Boston University economist Pascual Restrepo formalized this dynamic in a 2019 paper published in the Journal of Economic Perspectives, describing two forces that move in opposite directions whenever automation arrives. The displacement effect removes tasks from a role and hands them to a machine. The reinstatement effect creates brand new tasks that didn’t exist before, tasks that specifically require a human because they involve managing, directing, or correcting the very automation that displaced the old ones.
What New Tasks Are Emerging for Executive Assistants Right Now
Reinstatement is visible directly in the job today. Managing an executive’s AI scheduling tool, reviewing and editing AI-drafted correspondence before it goes out under the executive’s name, deciding which AI-generated meeting summary needs a follow-up action and which one doesn’t- none of these tasks existed five years ago. Acemoglu and Restrepo’s framework predicts exactly this pattern: the job doesn’t shrink to nothing; it becomes a different mix of tasks, weighted more heavily toward oversight and judgment and less toward mechanical execution.
Which Executive Assistant Tasks Are Actually Being Automated Today?
Scheduling, Drafting, and Summarization
Generative AI tools, including general-purpose assistants like ChatGPT and Claude alongside dedicated scheduling tools, now handle a meaningful share of routine drafting and calendar coordination reliably: a first-pass email response, a bullet-point summary of a long thread, a suggested meeting time based on stated constraints. These are precisely the codifiable, rule-based tasks the Frey and Osborne model flagged as automatable back in 2013.
What AI Still Can’t Do: Judgment, Discretion, and Representation
None of these tools decide whether a client’s “urgent” request actually deserves same-day attention, none of them handle a confidential conversation about why a meeting needs to be quietly rescheduled, and none of them represent an executive’s position in a live negotiation. That gap isn’t a temporary limitation waiting for the next model update. It’s a structural one: judgment calls require context, relationships, and accountability that a tool trained on patterns across millions of documents doesn’t hold about your specific business and your specific stakeholders.
What Does the World Economic Forum Say About the Future of Administrative Roles?
The Shift From Task Execution to Human-Machine Collaboration
The World Economic Forum’s ongoing research into the future of jobs has consistently framed the coming years not as humans being replaced by machines, but as a shift toward human-machine collaboration, where the most resilient roles are the ones that pair a person’s judgment with a tool’s speed rather than competing against it directly. Administrative and executive support work sits squarely in the category the Forum’s research flags as being redefined by this collaboration rather than eliminated by it, echoing the same displacement-and-reinstatement pattern Acemoglu and Restrepo describe in more technical economic terms.
How the Executive Assistant Role Is Already Changing
From Gatekeeper to AI Orchestrator
The traditional description of an executive assistant as a gatekeeper, filtering what reaches the executive, is expanding into something closer to an AI orchestrator: someone who sets up the scheduling tool’s rules, reviews what the AI drafts before it goes out, and decides when a task is safe to hand to automation versus when it needs a human touch specifically because the stakes are too high for an algorithm’s judgment.
New Skills Employers Are Screening For
Job postings for executive assistants increasingly mention comfort with AI tools directly, not as a bonus skill but as a baseline expectation, alongside the traditional core of discretion and communication. This isn’t replacing the traditional skill set; it’s adding a layer on top of it: the assistant who can direct three AI tools well and still make the judgment calls those tools can’t make is worth more, not less, than the assistant who ignores the tools entirely.
What This Means If You’re Hiring an Executive Assistant Today
Hire for the Skills That Don’t Automate
Frey and Osborne’s research, Bessen’s ATM study, and Acemoglu and Restrepo’s task model all point in the same direction for a hiring decision: screen hardest for the parts of the job that have consistently resisted automation, judgment under ambiguity, discretion with confidential information, and the relationship management a machine has no stake in. These are exactly the skills covered in a proper vetting process, not the software checklist most job postings lead with.
Build AI Fluency Into the Job Description, Not Around It
At the same time, treat AI tool fluency as a real, current-generation requirement rather than an optional extra. A virtual executive assistant who already knows how to set up and manage scheduling automation, review AI-drafted correspondence critically, and know exactly where the tool’s judgment stops. Their work begins by doing the job the way it actually looks today, not the way it looked before these tools existed. For businesses handling regulated or highly confidential work, an executive assistant for law firms needs this same AI fluency paired with an even higher bar for discretion, since the judgment calls AI can’t make carry more weight in that context.
The Oxford study that started this conversation over a decade ago was right about one thing: a large share of this role’s mechanical tasks really would become automatable. It was wrong about what that meant for the job itself, and the ATM story, the task-reinstatement research, and the World Economic Forum’s own framing all explain exactly why. Book a discovery call to talk through what the role looks like for your business today: judgment and discretion at the core, AI fluency layered on top, not one replacing the other.