A deep agent is an AI agent designed for complex, open-ended, long-horizon tasks. It typically combines planning, tool use, sub-agents, filesystem or artifact access, memory, and context management so it can make progress across many steps rather than answering in one turn.
The term is often used for agents that include:
- a planning or task-tracking mechanism;
- sub-agents for delegated work;
- a filesystem, workspace, or artifact store for context offloading;
- agent memory;
- durable execution or resumability;
- agent skills; and
- tracing and evaluation for long-running behavior.
A deep agent is not just a larger model. It is an agent design pattern or harness for sustained work. A smaller model with strong scaffolding, file access, and reliable orchestration may behave more like a deep agent than a frontier model wrapped in a single chat prompt.
The main risks are cost, runaway execution, stale state, tool misuse, and hidden failure accumulation. Deep agents need agent policy, sandboxing, human approval boundaries, and agent evaluation that checks both final outcomes and intermediate behavior.
LangChain describes Deep Agents as agents that can plan, use subagents, and use file systems for complex tasks in its Deep Agents overview.
The LLM Knowledge Base is a collection of bite-sized explanations for commonly used terms and abbreviations related to Large Language Models and Generative AI.
It's an educational resource that helps you stay up-to-date with the latest developments in AI research and its applications.
Previous