Harrison Chase's Mental Model: Agents = Model + Harness + Context

In his Sequoia talk "Owning Your Intelligence," LangChain founder Harrison Chase laid out a lean mental model for agents: Agents = model + harness + context. His central claim: owning your intelligence means more than downloading model weights—the harness and context matter just as much.

Confirmed

• Chase's formula has three components: model, harness (the orchestration/architectural machinery), and context; he also notes that an agent is made up of a system prompt, planning tools, a file system, and sub-agents.
• @dbreunig framed the harness as "Situated Agents," citing moves by major players like Databricks' Omnigent into embodied agents, and argued that the Agent Harness concept is on the rise.
• @hugobowne pointed out that the same foundation model under different constraints (harnesses) can behave as three distinct kinds of agents: a coding agent can inspect a repo, edit files, run tests, recover from errors, and work continuously for an hour, while a customer service agent exhibits entirely different behavior patterns under its own constraints.
• Building on Chase's ideas, @dfinke offered a set of formal extensions: first, the Agent (Harness) architecture = Model + Context + Tools + Control Loop; second, Behavior = Model + Context + Tools + Harness, emphasizing that model capability, context, external tools, and orchestration together determine agent behavior; and third, Behavior = Harness + Intent, underscoring the driving role of intent in the system.

Why it matters

• This mental model offers a unified framework for understanding today's agent products: as model capabilities converge, the harness and context become key differentiators—explaining how the same model can power wildly different products like coding and customer service agents.
• Chase's stance as LangChain founder, corroborated by dbreunig's observations of big-player moves, signals that the harness has evolved from a community concept into a core topic of industrial engineering practice.

2026-08-17 ~ 2026-08-18 · 8 related posts

• Same model, different harnesses: Coding vs. Support vs. Free Agents — hugobowne · 2026-08-17
• [source] Deep Dive: Harnesses are Situated Agents — dbreunig · 2026-08-17
• Agent Harness Trend: Situated Agents and the Rise of Coding Environments — dbreunig · 2026-08-17
• [source] Harrison Chase: Owning Your Intelligence with Agents — hwchase17 · 2026-08-18
• Formula for Agent Behavior: Model + Context + Tools + Harness — dfinke · 2026-08-18
• [source] Agent Essence Formula: Model + Context + Tools + Control Loop — dfinke · 2026-08-18
• Behavior Formula: Agent Architecture + Intent — dfinke · 2026-08-18

1 near-duplicate retellings: hwchase17