Is a Secretary job safe from AI?

Only partly. A Secretary scores 67/100 (medium risk). AI can already handle a meaningful share of the tasks, so the safest path is shifting toward the judgment, relationship, and oversight parts of the role.

When will AI be able to do a Secretary's job?

Based on AI's measured task-completion horizon (METR, doubling ~every 4.3 months), AI reaches this role's core complexity ~10 months. That projects the capability — actual adoption usually lags it.

How accurate is the Secretary AI-risk score?

It's a transparent, computed estimate — directionally useful, not a guarantee. It blends six labor and AI-exposure signals (O*NET, BLS, Eloundou task exposure, AIOE, the Anthropic Economic Index, and physical-automation data). See the methodology page for the full formula.

A Secretary carries a 67/100 AI replacement risk (medium). AI can already handle data entry and record-keeping; Resolving exceptions still needs a person. Of exposed work, ~61% is automation vs 39% augmentation. Capability clock: ~10 months. (ReplacedYet AI-Risk Index, 2026 data.)

Will AI replace a Secretary?

AI replacement risk: 67/100 (medium risk). Moderate exposure — AI automates routine parts; judgment and relationships remain human.

Timeline: 2028–2031. Of the exposed work, roughly 61% is likely to be automated and 39% augmented. $53.1B/yr of US wages sit in highly-exposed work for this role.

AI/software exposure: 75%. Robot/physical-automation exposure: 3%.

Capability clock: AI's measured task horizon reaches this role's core complexity (~27.8h of human work) ~10 months — projected from METR's ~4.3-month doubling.

Pressure Index: 73/100 (high) — blends risk, demand trend, and real-world evidence. Job postings down 38% vs 2020.

AI tools targeting this role

Microsoft Copilot — drafting documents, emails, and summaries
UiPath — automating repetitive back-office workflows

Layoff signal: high — Routine administrative tasks are highly automatable, and many organizations have cited AI in trimming support roles.

Tasks at risk

Data entry and record-keeping — OCR + LLMs capture and structure documents end-to-end.
Form processing and validation — Rule and model-based checks run without a person.
Routing and filing information — Software moves and files records automatically.

Tasks that still need a human

Resolving exceptions — Messy, ambiguous cases still escalate to a person.
Coordinating with people — Chasing missing info and coordinating stays human.

Skills that protect you

Data quality & analytics — Govern and analyze data instead of just entering it.
Automation / RPA tooling — Operate the systems that replace manual work.
Process ownership — Own the workflow, not just a step in it.

Safer adjacent careers

Office Manager (26%) · Childcare Worker (5%) · Hairdresser (6%) · Nursing Assistant (6%)

Frequently asked questions

Will AI replace Secretarys?: A Secretary carries a 67/100 AI replacement risk (medium). AI can already handle data entry and record-keeping; Resolving exceptions still needs a person. Of exposed work, ~61% is automation vs 39% augmentation. Capability clock: ~10 months. (ReplacedYet AI-Risk Index, 2026 data.)
Is a Secretary job safe from AI?: Only partly. A Secretary scores 67/100 (medium risk). AI can already handle a meaningful share of the tasks, so the safest path is shifting toward the judgment, relationship, and oversight parts of the role.
When will AI be able to do a Secretary's job?: Based on AI's measured task-completion horizon (METR, doubling ~every 4.3 months), AI reaches this role's core complexity ~10 months. That projects the capability — actual adoption usually lags it.
How accurate is the Secretary AI-risk score?: It's a transparent, computed estimate — directionally useful, not a guarantee. It blends six labor and AI-exposure signals (O*NET, BLS, Eloundou task exposure, AIOE, the Anthropic Economic Index, and physical-automation data). See the methodology page for the full formula.

Category: Administrative · Methodology · Download the dataset

ReplacedYet AI-Risk Index. Last updated 2026-06-27. AI-estimated and directionally useful, not a guarantee.