Nature: Human Scientists Still Outperform the Best AI Agents on Complex Tasks

Nature: Human Scientists Trounce the Best AI Agents on Complex Tasks

Nature published a piece on April 13, 2026 titled 'Human scientists trounce the best AI agents on complex tasks.' The piece compares expert human scientists with frontier AI agents on long, multi-step research tasks.

What the Piece Says

According to Nature, expert humans still outperform the best AI agents on the hardest benchmark tasks. The gap is largest on problems that require sustained reasoning, planning, and judgment about which approach to drop.

Nature notes that AI agents do better than humans on some retrieval-heavy and coding-style tasks. The piece frames the human-AI gap as narrowing in some areas and still wide in others.

A Note on Exact Scores

Specific numerical scores for individual models have circulated in follow-up coverage. Readers who want the exact benchmark breakdown should check the primary paper and its supplementary tables rather than secondary summaries.

Why It Matters

Many production AI benchmarks are close to saturated, yet on harder research tasks, frontier models are still below expert human performance. For teams deploying AI agents, the piece is a reminder that results on standard leaderboards do not automatically transfer to the open-ended research work a senior expert handles.