←Home

    The Research Agent That Keeps Going

    How AI research agents are being built to sustain long-running investigations, carry memory across sessions, learn from their own experiments, and even rewrite their own research code, plus how we can tell whether those runs actually reach sound conclusions.

    In order

    1 stories

    Photo: Jordyn St John

    Discovery · Metacognitive research agents

    An AI research agent that keeps notes on what it knows before deciding what to try next

    Shanghai AI Laboratory's Sage-Taiyi adds a supervisory layer that tracks findings, assumptions and open questions, then uses them to steer an AI's scientific idea search.

    arXiv preprint · 9 November 2026