MIT study: AI works like a 'disillusioned intern' — good enough, but not outstanding
What it really says
MIT researchers tested 41 AI models (Claude, Gemini, ChatGPT) on over 11,000 work tasks. Result: AI achieves 'minimally sufficient' quality (grade 7/10) in 65% of tasks. For higher requirements (grade 9), the success rate drops below 50%. Researchers compare current AI to a 'disillusioned intern' — minimum standards yes, outstanding work no.
Our assessment
This is one of the most important studies against panic narratives. Yes, AI can do many things — but 'good enough' is not 'replaces the professional'. Especially for tasks requiring creativity, empathy, critical thinking, or contextual knowledge, AI remains clearly behind human quality. The study shows: AI is a tool, not a replacement.
Relevance for Germany
Reassuring news for everyone worried about their job. MIT data contradicts headlines about mass unemployment. Especially relevant for the German economy with its high proportion of specialized professionals.
Fact check
Peer-reviewed MIT study with 41 models and 11,000+ tasks. Methodology was rated solid by independent researchers. The 65% figure refers to 'minimally sufficient' (score 7/10).
Source
- • MIT CSAIL Working Paper 04/2026
- • Axios Reporting 02.04.2026
- • Fortune Analysis 03.04.2026