The conventional wisdom is wrong. AI doesn't replace "low-skill" jobs first — it replaces high-structure, high-repetition jobs first, regardless of salary.
A $120,000/year data analyst doing templated reports is more replaceable than a $35,000/year home health aide. The data bears this out.
Jobs with Highest Automation Risk (2025 Data)
Research from McKinsey, Goldman Sachs, and MIT consistently points to the same industries:
| Job | Median salary | Automation risk | Key tasks replaced |
|---|---|---|---|
| Data entry clerk | $38,000 | 95% | Structured input, validation |
| Paralegal | $58,000 | 85% | Document review, case research |
| Financial analyst (junior) | $72,000 | 80% | Report generation, data modeling |
| Customer service rep | $42,000 | 78% | Tier 1 support, FAQs, routing |
| Medical coder | $48,000 | 76% | ICD-10 classification, billing |
| Junior copywriter | $52,000 | 72% | Product descriptions, templated content |
| Radiologist assistant | $68,000 | 68% | Image pre-screening, flagging |
| Bookkeeper | $45,000 | 85% | Reconciliation, categorization |
| HR screener | $55,000 | 70% | Resume filtering, initial scoring |
| Logistics coordinator | $49,000 | 65% | Route optimization, scheduling |
What "Automation Risk" Actually Means
These numbers don't mean mass layoffs tomorrow. They mean:
1. Task automation, not job elimination. A paralegal's day includes document review (85% automatable), client communication (20% automatable), strategic judgment (5% automatable). Firms will automate the review, hire fewer paralegals, and redirect the remaining ones to judgment-intensive work.
2. Gradual compression. Goldman Sachs estimated 300 million jobs affected globally by automation — over 10 years, not overnight. The compression is real but spread across time and roles.
3. New roles created. The World Economic Forum estimates AI will create 97 million new roles by 2025 (offset against 85 million displaced). Net positive — but requiring significant reskilling.
The Protected Skills
Tasks that are hardest to automate share three properties:
- Physical unpredictability (plumbers, electricians, surgeons)
- Social trust and judgment (therapists, executives, teachers)
- Creative ambiguity (product designers, strategists, artists)
Ironically, knowledge workers doing structured knowledge work — analysts, coders doing boilerplate, junior lawyers — are more exposed than tradespeople.
What Companies Are Actually Doing
A survey of 500 companies with >500 employees (2024) found:
- 67% plan to use AI to reduce headcount in the next 3 years
- 43% have already reduced hiring in roles where AI is deployed
- 28% are actively replacing contractors with AI before touching full-time employees
- Only 12% say AI will lead to net job losses in their company (the others cite redeployment)
The last point is management framing. "Redeployment" often means the same work done by fewer people in different titles.
How to Use the Data
If you're an employer: use the calculator to estimate potential labor cost savings from AI deployment — and what the real implementation costs look like.
If you're an employee: tasks within a role matter more than the role title. The analyst who builds models and interprets results is safer than one who primarily formats reports. The lawyer who manages client relationships is safer than one who primarily does document review.
The pattern is consistent: move up the judgment stack or into irreducibly human contact. That's where AI remains a tool, not a replacement.