new·The score now tells you which way it movedA brain's exam only ever grows: its own material writes questions, and so does every question a real caller asked and did not get answered. The score is a percentage over that growing set, so a brain that learned more could post a smaller number — and this week three did. One of them answered two MORE questions than the week before and showed eighteen points less. Printed as a single percentage, that reads as decline to a reader and as punishment to anyone who contributes material.all news →
mozg.beta
Sign in

LangChain · Agents · all subjects

agents/deep-agents

5 notes, read out of this brain and free to use. Each one was extracted from a source and is re-checked against its exam.

Deep Agents pre-assembles common agent stack

create_deep_agent pre-assembles the agent harness stack for long-running coding and research tasks with commonly useful capabilities already assembled, such as planning, file system tools, subagents, memory, and prompt caching. Use create_agent when you need to configure the harness yourself.

Deep Agents vs create_agent differences

Both create_agent and create_deep_agent provide fine-grained control over tools, memory, and more. The main difference is that Deep Agents comes with commonly useful capabilities already built in, such as planning, file system tools, and subagents. If the Deep Agents default harness does not fit your needs, you can start with create_agent and assemble the harness one piece at a time.

Build data analysis agent step progression

A data analysis agent can be built incrementally: (1) Minimal agent with baseline loop (model + tools, no harness); (2) Sandbox + filesystem for reading CSVs and running Python in isolated backend with file and execute tools; (3) Summarization middleware for automatic history compression when long sessions hit context limits; (4) Skills middleware for on-demand domain expertise via progressive disclosure; (5) Subagent with TodoListMiddleware for isolated visualization work and parallel delegation.

Deep agent middleware composition

The deep agent is composed of: FilesystemMiddleware + LangSmithSandbox for isolated filesystem and execute tool; SummarizationMiddleware for automatic context compression; SkillsMiddleware for on-demand domain knowledge; TodoListMiddleware + SubAgentMiddleware for parallel visualization subagent. This composition is the same foundation as create_deep_agent, assembled manually for control over what is included.

LangSmith setup for tutorial prerequisites

The deep agent tutorial requires LangSmith API key setup: (1) Sign up for a free account at https://smith.langchain.com using Google, GitHub, or email; (2) Create an API key in Settings → API Keys; (3) Export LANGSMITH_API_KEY environment variable; (4) Enable tracing by exporting LANGSMITH_TRACING=true to inspect tool calls, middleware steps, and subagent delegation.

Give your agent this brain