Chroma | Context Engineering Episode 3 - Lance Martin - LangChain

Описание: Chroma CEO Jeff Huber sits down with Lance Martin to discuss the current state of agents and more.

Find Lance on X, https://x.com/RLanceMartin, and his website, https://rlancemartin.github.io/

0:00 Introduction & Welcome
0:09 Context Engineering: What It Is and Why It Matters
2:05 Context Rot and Performance Degradation
3:31 Year in Review: 2025 AI Trends
4:28 Giving Agents a Computer (File System & Shell)
5:00 Model Context Protocol (MCP) and Tool Bloat
6:07 Multi-Tier Action Space Architecture
8:24 Tool Search and Progressive Disclosure
10:33 Agent Harness Structure & Deep Agents
12:17 Skills and Standard Operating Procedures (SOPs)
14:13 Context Offloading Techniques
15:49 Plan Offloading & The Ralph Wiggum Loop
18:00 Context Caching for Cost & Speed
18:27 Sub-agents and Context Isolation
21:16 Summary: Key Context Engineering Principles
22:00 Evolving Context & Continual Learning
25:02 Claude Diary: Reflecting on Sessions
26:06 Skill Learning from Agent Trajectories
27:00 Memory Management in Token Space vs Weights
28:35 RLMs: Reason Language Models & Learned Context Management
31:30 What Can Be Absorbed Into Models (The Classifier Test)
35:30 Memory: Writing vs Retrieval Challenges
40:00 File Systems as Agent Primitives
42:46 Limitations of File Systems for Large Codebases
45:02 Multi-Agent Collaboration & Concurrency Challenges
49:35 Layers of Context: Session, Agent, Organizational, Global
52:27 File Systems vs Databases: A Hot Take
55:51 Sandboxing and Agent Infrastructure
58:42 What's Most Exciting: Memory, Personal Agents & Bioscience
2:02:18 Wrap Up

—

Chroma is the open-source AI application database. Batteries included.

Embeddings, vector search, document storage, full-text search, metadata filtering, and multi-modal. All in one place. Retrieval that just works. As it should be.

Try it today:
https://trychroma.com/cloud