LangChain ships AI Search, Red Teaming, and fine-tuning across LangSmith update
LangChain · x · 2026-10-19
LangChain's October newsletter rounds up everything shipped at Interrupt NYC & London, themed around helping teams "own their intelligence" — the model, harness, and context.
New in LangSmith:
- AI Search: semantically search traces with questions like "did the user get frustrated?" or "did PII leak?"
- Managed Deep Agents v0.9: context-aware agents with Slack reactions, user-level memory, scheduled tasks, and free built-in web search
- Engine v2: Red Teaming plus detection of inefficient agent work and metric trends, with automatic fix validation
- Trajectories: chronological view across main agent and subagents
- Custom Apps: build and run custom interfaces on your LangSmith data
- Fine-Tuning: turn production trajectories into training data, tune via Baseten or Fireworks, evaluate and serve
Open source: LangChain v1.4.0 supports FastMCP via the langchain.mcp namespace.
More from coding & agent
- Build Systems, Not Just Prompts: The Fastest Way To Learn AI, According To EXM7777 — EXM7777 · 2026-10-19
- Sequoia backs Blacksmith to verify code as fast as agents write it — Sequoia Capital · 2026-10-19
- Engineer open-sources his "software factory": a modular, harness-agnostic agent skill workflow — JnBrymn · 2026-10-19
- Dev proposes rules for AI email agents: disclose identity, separate inbox, keep humans cc'd — curious_vii · 2026-10-19
- Designer: Claude builds interactive websites in 30 minutes that took a team 1-2 weeks — Tegadesigns · 2026-10-19
- Running a Jev decision-model LoRA to play snake: 600ms per move, 2200+ ticks — BitGreen1270 · 2026-10-19