100-Page Survey on Long-Horizon Agents
青稞AI · wechat · 2026-07-18
Taking three months to compile, the 100-page survey "Towards Long-Horizon Agents: A Survey" systematically outlines the development trajectory and key technologies of long-horizon agents.
The paper frames "long-horizon capabilities" as the co-evolution of Harness Engineering and Model Optimization: explicit harness capabilities will gradually be internalized into model policies, and stronger policies will, in turn, drive more powerful harnesses.
The survey explores six main aspects:
- Foundations: Definition and difficulty grading of long-horizon capabilities
- Evolution: From prompt engineering to context engineering, and then to harness engineering
- Harnesses: Loops and workflows, context and memory, tools/MCP and skills, orchestration, hooks, verification
- Optimization: Architecture, data/environment synthesis, pre-training/mid-training, fine-tuning, reinforcement learning, policy distillation, self-evolution
- Applications: Software engineering, information retrieval, Computer Use, multimodality, general agents
- Frontiers: Evolution, effectiveness, efficiency, and trustworthiness
Repository and homepage links are included for further in-depth reading and tracking.
More from coding & agent
- Soft Clamp cuts tool-call overuse in multi-teacher distillation, from 13.7% to 9.0% — antgroup · 2026-07-21
- Agent harness memory loss and compaction are still a major usability problem — adityaag · 2026-07-21
- SpecJudge runs locally on Ollama to pick the right-sized AI model for your project — jokiruiz · 2026-07-21
- A developer maps out six design rules for CLIs that humans and AI agents can both use — yujiezha · 2026-07-21
- A coding-agent skill that forces ADHD-friendly, answer-first output — ayghri · 2026-07-21
- A set of agent skills for CAD, robotics, and hardware design — earthtojake · 2026-07-21