Context Engineering for Agents - Lance Martin, LangChain
- Source
- youtube.com
- Author
- Latent Space
- Date
Context
Lance Martin of LangChain says in a Latent Space podcast that an is essentially tool calls in a loop, and each call's output flows back into the model's . Building his open deep research agent, he says a naive version used about 500,000 (chunks of text the model reads and writes), roughly one to two dollars a run. He also points to Chroma's report on context rot, which describes performance falling as context grows.
His answer is . One method is offloading: write bulky results to a file or the agent's state, and return a summary or reference the agent can fetch later. He says the summaries must be prompted carefully for recall so the agent knows what each stored item contains. These are a practitioner's experience and cited reports, not a controlled comparison.
- 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.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- 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.
Checking sign-in…
Loading comments…





