Chroma | Context Engineering Lance Martin - LangChain
17 snips
Jan 16, 2026 
Join Lance Martin, an AI practitioner known for his expertise in context engineering and LangChain, as he dives into fascinating topics. He discusses the importance of context engineering and how to manage it with tools like shells and file systems. Lance explains context rot, model context protocols, and the challenges of memory management in AI systems. Delve into subagents for task isolation and the nuances of multi-agent collaboration. Get insights on evolving context and the future of personal assistant agents!
AI Snips
Chapters
Books
Transcript
Episode notes
Isolate Tasks With Sub-Agents And Progress Files
- Use sub-agents to isolate context for parallelizable or long-running tasks and communicate via files or progress logs.
- Structure long tasks as loops where each agent reads progress, performs a task, and writes updates to the file system.
Continual Learning In Token Space Is Practical
- Evolving context (continual learning) can live in token space as editable external artifacts like skills, prompts, and diaries.
- These external, human-interpretable artifacts offer easy experimentation without weight updates.
Let Models Learn Context Management
- Train models (RLMs) to manage context: when to offload, spawn sub-agents, or evolve memories based on trajectories.
- Use learned context managers to automate rules that are verifiable and classifier-friendly.
