LangChain ships Engine v2, Managed Deep Agents v0.8 and fine-tuning at Interrupt NYC
LangChain · x · 2026-09-26
At its Interrupt NYC conference, LangChain announced a batch of LangSmith updates organized around two pillars of the agent development lifecycle—runtime, and observability/evals:
- LangSmith Engine v2: adds red teaming, expanded issue detection, and automated fix validation to catch agent problems proactively.
- Managed Deep Agents v0.8: now includes user-level memory, built-in web search, and identity-scoped auth for shipping production agents.
- Trajectories: a new conversational view of traces that renders agent sessions as dialogues.
- LangSmith Fine-Tuning: uses trajectory data to fine-tune models for specialized agent tasks.
- Custom Apps: lets teams build and publish custom LangSmith interfaces for workflows like annotation queues and experiments.
The overall pitch is a closed loop: runtime (memory, tools, secure execution) plus observability, evals, and fine-tuning compound agent intelligence as usage grows.
Related event: LangChain Unveils LangSmith Engine v2 and Suite of Agent Updates(3 posts)→
More from coding & agent
- Reasonable Team Publishes TLA+ Tutorial: Not a Silver Bullet, AI Agents Could Change That — fhuszar · 2026-09-26
- Using Opus 5.5 to make a video explaining Algorithm W and Hindley-Milner type inference — ctjlewis · 2026-09-26
- Runway MCP brings Gen-4.5 video and image generation into Claude chats — runwayml · 2026-09-26
- One 'Harmless' Agent Permission Turned Into a Backdoor for Everyone — Adorable-Algae6903 · 2026-09-26
- DEV·TV: a single-HTML-file TV playing Hugging Face models and AI papers hits Product Hunt — TradingCardGirl · 2026-09-26
- MCP Wallet Agents Need Boring Permission Rules Before Going Agentic — Any-Sir-7622 · 2026-09-26