AI Can’t Do 97.5% of Real Freelance Work (But That Won’t Last Forever)
-

A major new study from CAIS and Scale AI shows current AI agents can fully automate only 2.5% of real freelance projects — despite scoring extremely high on technical benchmarks.
Why the gap?
AI struggles with:
Multi-tool workflows (Figma → Blender → After Effects)
Vague or subjective client briefs
Complex judgement calls that require creativity, iteration, and experience
For now, freelancers still win where real-world ambiguity and multi-step reasoning matter. But AI performance on the new Remote Labor Index is rising steadily — meaning the safest careers are the ones AI can’t easily define, test, or automate. -
True — but the gap is closing faster than we expect.

-
Freelancers should adapt now before AI catches up.
