We Researched 88 Jobs, One at a Time, to See Which Ones AI Is Actually Replacing

Most of the AI-and-jobs debate assumes a binary -- safe or replaced. Our own occupation-by-occupation research says most jobs are doing neither.

We didn't survey anyone or run a model that spits out a percentage. We went occupation by occupation -- 88 in total -- and wrote out, role by role, what's automated today and what still needs a person, with the reasoning published alongside each one. Sorted into three tiers, the shape is: 15 roles are high exposure, meaning the core of the job that used to take up most of the week is substantially automatable right now. 51 roles are moderate exposure, where the job is splitting into an automating half and a growing half, and the ratio depends a lot on who's actually doing it. 22 roles are low exposure, for reasons that don't look like they're eroding soon.

That middle number is the one people miss. Fifty-one out of eighty-eight is the largest tier by far, and it doesn't fit the safe-or-replaced framing most coverage defaults to. Two people holding the exact same job title can sit at very different points in that split -- a marketing manager who spends most of the week drafting copy and building decks is in a different position than one who spends most of it on stakeholder negotiation and budget calls, even though their LinkedIn headline reads identically.

The high-exposure tier is where the mechanical core of a job -- the part that's repetitive, rule-based, and has a checkable right answer -- has genuinely been replaced. Transcription and translation are furthest along. Data entry, bookkeeping, administrative assistance, copywriting, proofreading, graphic design, IT support tickets, data analysis, customer service replies, paralegal document review, QA testing, and technical writing round out the rest of that tier, all for the same underlying reason. Take data analysts, since we get asked about that one constantly: writing a SQL query or building a recurring dashboard is close to instant now. What's left is knowing which question was worth asking, and getting a skeptical stakeholder to act on the answer -- arguably the only part of the job that was ever the hard part.

Low exposure doesn't mean AI-proof forever, either. It means the structural reasons a role resists automation -- physical dexterity in unpredictable settings, high-stakes accountability, deep trust relationships -- are fully intact today, for electricians, dentists, chefs, and similar roles, and show no near-term sign of eroding. That's a different kind of durability than "nobody's built the tool yet," but it's not a permanent guarantee either.

None of this sits in a vacuum. Tech companies alone cut roughly 170,000 jobs in 2026, and white-collar payrolls -- professional services, financial roles, information and media -- have contracted for 31 straight months, a streak that historically only shows up around recessions. Overlay that against our tiers and the sectors seeing the heaviest 2026 cuts map closely onto our moderate and high-exposure occupations. That's what the automating half of a job leaving actually looks like from the outside, once you strip away the press release language around it.

Job Resiliens' own task-by-task AI Job Risk Index research across 88 occupations (jobresiliens.com/research/ai-job-risk-index), cross-referenced with 2026 tech-sector layoff tracking and U.S. Bureau of Labor Statistics payroll data.

Also worth reading: AI vulnerability score by job title · how we score AI resilience