Agents cut in half

·2026.04.15 00:03

Key point

In Stanford's 2026 AI Index, AI agents scored only about half the level of PhD experts.

In Stanford's 2026 AI Index, AI agents performance was shown to be at half the level of the PhD experts benchmark.

The core message is that agents still have a significant gap compared to human experts on high-difficulty professional tasks.

Comparison: agents vs. PhD experts
Result: agent scores are at about 50% of expert level
Implication: while automation potential has grown, clear limitations remain in high-difficulty judgment and reasoning

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.