LangChain AI Updates: October 29, 2026

1. LangChain Adds a Schedules SDK to Managed Deep Agents for Proactive, Cron-Driven Agents

LangChain published a post by Nathan Drezner arguing that most proactive, always-on agents run on a simple mechanism: agent runs triggered as cron jobs. The post lists what scheduled runs make possible: memory consolidation (“dreaming”) on a daily cadence, goal-driven proactive messages, triage of Slack or email streams, user-requested follow-ups, and orchestrator agents that wake themselves to check on async subtasks. LangChain says more than 70% of Managed Deep Agents runs on LangSmith now come from schedules rather than a human message. Managed Deep Agents, hosted on LangSmith Deployments, include a built-in cron server. Schedules can be static, defined with define_schedule in a schedules/ folder, or created at runtime by wrapping the Schedules SDK (runtime.schedules.create, list, delete) in agent tools, including one-off runs at a set time. Each scheduled run uses the thread and memory of the principal who created it. Source

2. OpenWiki Plus an Evidence Graph Makes Agent Memory 5x More Accurate on Cross-Document Questions

LangChain tested OpenWiki, its open-source wiki compiler built on Deep Agents, as a memory layer for a retrieval agent over a synthetic insurance library of 111 documents and a ledger of 972 facts. On 50 questions that require following dependencies across documents, accuracy rose from 0.10 with keyword search alone to 0.52 with the compiled wiki, and cost per correct answer fell from about $0.55 to $0.19. On 80 simple lookups the wiki made no difference to accuracy and added about 3 seconds of latency. Compiling the library cost $4.15, and adding documents later cost about $0.09 each. Because citation consistency across runs stayed low, the team added an evidence graph that turns existing wiki citations into typed edges (cites, refers_to, supersedes, amends) between claims, pages and paragraphs, built in under a second with no model calls. Navigated through locate and expand tools, the graph improved accuracy by a further 27%, doubled citation consistency, and cut cost per correct answer by 58% versus the wiki alone. Source