type: decision
status: active
timestamp: 2026-07-02
tags: [stack, ai, orchestration, langchain, langsmith, langgraph, deferred]

LangChain ecosystem — deferred, revisit 2026-10-02

LangChain / LangGraph / LangSmith / integrations. Not adopted, not rejected. Current MCP + skills + AGENTS.md stack covers the same surface. Revisit in 3 months if real gaps surface.

LangChain — deferred

Decision

Not adopting LangChain, LangGraph, LangSmith, or integrations layer. Not rejecting either — set a review checkpoint 2026-10-02.

Why deferred (not rejected outright)

The pitch — chains + tools + memory + tracing — maps 1:1 to what MCP + skills + knowledge/ + Claude Code’s built-in traces already deliver. But:

  1. Multi-agent orchestration is evolving fast. LangGraph’s graph-based branching MIGHT become useful when subagent orchestration outgrows what Agent tool + pipeline() primitives cover.
  2. LangSmith’s structured eval framework has no direct MCP equivalent yet. If eval discipline matures, it may fill a real gap.
  3. The no-rebuilding-free-software rule cuts both ways: adopting LangChain is renting a rebuild of your stack; but skipping observability tools is rebuilding LangSmith yourself with grep-through-transcripts.

Why not adopt now

Review checkpoint

Revisit 2026-10-02. Trigger a re-evaluation if:

If none of those, extend deferral another 3 months.

Cross-refs


Edit on GitHub · Back to index