Writing
Notes from inside agent harnesses
Long-form pieces on how agent harnesses work, covering the loop, middleware, context, delegation, storage and control. New parts every three or four days.
- 14 Nov 2026 · Deep Agents, from the inside, part 3
How a tool call works, and one turn of the loop
Tool calling from first principles. What the model actually returns, how LangChain turns it into a graph, and the six middleware hooks every Deep Agents feature plugs into.
- 13 Nov 2026 · Deep Agents, from the inside, part 2
Your first deep agent, and the prompt it really sees
Install Deep Agents, run a first agent, read its files back, stream subagents and pause for approval. Then every create_deep_agent() argument, and how the system prompt is assembled layer by layer.
- 12 Nov 2026 · Deep Agents, from the inside, part 1
What Deep Agents is, and where it came from
LangChain's Deep Agents is a harness, not a runtime. Where it sits in the stack, what it adds over a plain agent loop, and the four ideas it started from.