LangChain Pitches Deep Agents as a Fix for Context Engineering
- Source
- x.com
- Date
The biggest challenge facing an agent harness is context engineering, or what information an agent gets, and when. The longer the task, the harder it is to keep the context window from overloading. Deep Agents ships with this built in: filesystems, subagents, and skills, already wired up for context management.
Context
LangChain frames the central difficulty in building an as , deciding what information an receives and when, since longer tasks make it easier for a model's to fill with irrelevant or stale information. LangChain's post says its Deep Agents framework ships with three primitives built in to manage that: a filesystem the agent can read and write to instead of holding everything in its context, subagents it can delegate narrower pieces of work to, and skills it can load only when relevant. LangChain has already built at least two applications on this framework: Managed Deep Agents, which runs every evaluation in a fresh container to catch regressions when a model or tool changes, and Deep Life Sci, a pharma research agent built on the same base that pulls from clinical trial records and published papers.
- agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
- context engineering — Deciding what an agent sees at each step — the discipline that replaced prompt engineering once context windows got large enough to fill badly.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…