LangChain Publishes Practical Guide to Building Custom Agent Harnesses
hwchase17 · x · 2026-09-14
LangChain engineer Sydney Runkle published a practical guide to building domain-specific agent harnesses. Key ideas: an agent is model + harness, where the harness is scaffolding that feeds the model the right context at every step; createagent is LangChain's primitive for custom harnesses with custom prompting, business logic and guardrails, complementing pre-assembled harnesses like Deep Agents and Claude Agent SDK that ship opinionated middleware (memory, context management, sandboxing). Includes runnable code examples, amplified by Garry Tan's quip: 'either you die a system of record or you live long enough to become a domain-specific harness.'
Related event: LangChain Publishes Guide to Building Domain-Specific Agent Harnesses(2 posts)→
More from coding & agent
- OpenViking launches as an open-source context database for AI agents — adnan_hashmi · 2026-09-14
- Git worktrees ~80% faster via APFS/btrfs/XFS/ReFS folder clones, with big disk savings — steipete · 2026-09-14
- Dev burns ~80% of a 20x AI sub debugging cascading AWS failures — generativist · 2026-09-14
- Microsoft's AgentRx paper pinpoints where long agent traces really fail — JohnAlexander · 2026-09-14
- Dev tip: route Claude Code subagents to Opus to save premium usage — lydiahallie · 2026-09-14
- Engineer maps NFL offense onto agent architecture: QB as main agent, WR as tools — philipkiely · 2026-09-14