deepagents

Overview

Deepagents is an open-source agent framework associated with Harrison Chase and the LangChain ecosystem. It was introduced in the newsletter as an open-source Claude Agent SDK for building and running Claude-powered agents, and later referenced as part of a broader agent stack that includes memory, MCPs/skills/subagents, triggers, and human-in-the-loop workflows. The concept appears to sit at the intersection of developer tooling, agent orchestration, and product infrastructure for teams building production-grade AI agents.

For AI Product Managers, Deepagents matters because it reflects a broader shift in agent platforms: from demo-oriented wrappers around frontier models to extensible frameworks that support customization, middleware, observability, and model portability. Newsletter mentions also suggest an important strategic direction—optimizing Deepagents for open-source models as closed-model costs rise—which makes it especially relevant for PMs evaluating long-term cost, control, and deployment flexibility.

Key Developments

• 2026-01-14: Deepagents was highlighted as a feature within the generally available LangSmith Agent Builder, alongside memory, MCPS/skills/subagents, triggers for autonomous workflows, and an agent inbox for human review.
• 2026-04-10: Harrison Chase noted that community middleware for customizing agents and Deepagents was emerging, citing langchain-task-steering as an example and inviting contributors to participate.
• 2026-04-13: Harrison Chase launched Deepagents as an open-source Claude Agent SDK at `github.com/langchain-ai/deepagents`, positioning it as a framework for building and running Claude-powered agents.
• 2026-04-30: Harrison Chase said he was optimizing Deepagents for strong performance on OSS models, based on the view that closed-model costs could become prohibitively high by 2026.

Relevance to AI PMs

1. Framework choice and roadmap planning: Deepagents is a signal that agent infrastructure decisions increasingly involve ecosystem fit, extensibility, and model portability—not just raw model quality. PMs can use this lens when deciding whether to build on proprietary SDKs or open frameworks. 2. Cost and vendor strategy: The shift toward OSS model optimization is directly relevant for PMs managing inference budgets, margin pressure, or enterprise deployment constraints. Deepagents suggests a path toward reducing dependency on expensive closed models while preserving agent capabilities. 3. Customization and platform design: The mention of middleware such as task steering indicates that agent products may need configurable orchestration layers. PMs should think about where customization belongs in their stack: prompts, routing logic, tools, memory, or reusable middleware.

Related

• Harrison Chase: The primary figure associated with Deepagents and its product direction.
• LangChain: Deepagents is connected to the LangChain ecosystem and appears aligned with its broader agent tooling strategy.
• LangSmith Agent Builder: Deepagents was referenced as part of the LangSmith Agent Builder feature set, suggesting integration with no-code or operational agent-building workflows.
• langchain-task-steering: An example of community middleware for customizing agent behavior in Deepagents.
• Memory: Mentioned alongside Deepagents as part of a broader agent system, indicating support for persistent context and state.
• MCPS/skills/subagents: Related building blocks that suggest modular agent composition and tool use.
• Anthropic / Claude: Deepagents was initially framed as a Claude Agent SDK, tying it to Anthropic’s model ecosystem.
• OSS models: A major strategic theme in later mentions, as Deepagents is being optimized for open-source model performance.