Theory of Agent survey from Beihang and 7 institutions reviews ~600 agent studies
jiqizhixin · x · 2026-09-24
Beihang University, SUSTech, Fudan, BIT, University of Edinburgh, BJUT, ICT CAS and Baidu present the Theory of Agent survey, organizing 600 recent studies through a unified lens of internalization and externalization.
- Core question: when should an agent think on its own vs call a tool—overacting causes errors and risks, overthinking with underacting yields fluent but unsupported answers.
- Framework covers the full agent lifecycle: design, learning, alignment, evolution, and evaluation.
- Key claim: stable, repeatedly used knowledge and procedures should be internalized into the model, reducing redundant retrieval/tool calls, lowering latency, and letting capabilities accumulate through training.
More from coding & agent
- Building Your Own Algorithmic Trading Analyzer with Claude Code — mdancho84 · 2026-09-24
- Rumor: OpenAI Plans a Staggering Number of Releases Today, New Agent Teased — iruletheworldmo · 2026-09-24
- DHH says hand-writing code is over as Pragmatic Engineer unpaywalls AI coding mega-trend piece — IgorCarron · 2026-09-24
- Claude Code on Opus 5.5 directs a full Dario Amodei biopic trailer end-to-end — henkvaness · 2026-09-24
- WebMCP Challenge entries: agent-to-agent interviews, DNA monster game, auto-posting tool — haltakov · 2026-09-24
- Geoffrey Litt shares where to store AI context: docs, skills, and memories — thesaraharminta · 2026-09-24