What it is
LangChain describes Deep Agents as an agent harness. Its documentation combines the tool-calling loop with planning, filesystem access, context management and subagents. Execution and persistence depend on the backends and permissions configured by the developer.
Compare it as a software foundation for building an agent, not as a ready-made hosted assistant. The application owner still needs to choose a model, configure tools, control access and evaluate the resulting workflow.
Documented capabilities
- Planning
- Context management
- Filesystem tools
- Subagents
These labels summarise the cited documentation. They are not an independent assessment of reliability or task success.
Questions to ask before choosing
- Which backend stores files and conversation state?
- Are tool and filesystem permissions explicit?
- How are model calls, latency and execution costs observed?
- Can the workflow recover after a failed tool call or interruption?
Understand the surrounding harness before treating a product comparison as a model comparison.