Deep Agents

Long-running autonomous agents: background execution, checkpoint-resume, human approval queues, and managing agents that run for hours or days.

Deep Agents is the third layer of the LangChain stack — an agent harness with built-in planning, virtual filesystem, subagent spawning, and context engineering. The bridge from create_agent to production-grade autonomous agents.

Context engineering is THE core challenge in agent building. Deep Agents provides automatic context offloading, summarization, isolation via subagents, and progressive disclosure — keeping agents effective over long-running tasks.

Custom tools, filesystem backends, planning strategies, middleware, subagent configuration, and deployment — everything you need to tailor Deep Agents for your production use case.

ACP standardizes communication between coding agents and code editors — expose any Deep Agent as a coding assistant in Zed, JetBrains, VS Code, and Neovim via a single protocol.

Fleet is LangSmith's no-code agent builder — create agents from templates, connect event-driven channels (Gmail, Slack, Teams), schedule recurring runs, and deploy with zero infrastructure.

The current state of AI coding agents: Claude Code, Cursor, GitHub Copilot, Deep Agents CLI, and OpenAI Codex — what each does, how they work, and when to use which.

Build a custom coding agent with Deep Agents: sandbox execution, project-specific skills, edit-verify loops, and ACP exposure for IDE integration.

Code generation prompting is different from general prompting: specify the language, framework, and style; provide context via file references; request tests alongside implementation; and constrain the output format.