Why Does LangChain’s New Jev Auto Mode Forget Your Request at Tool Call 16?
Last Updated on September 22, 2026 by Editorial Team
Author(s): Chew Loong Nian – AI ENGINEER
Originally published on Towards AI.
Why Does LangChain's New Jev Auto Mode Forget Your Request at Tool Call 16?
The article explains that LangChain’s Jev-powered Auto Mode, as implemented in the current middleware, starts “going blind” mid-run: from tool call 16 onward, the guard asks whether an action is user-authorized without including the user’s message in the classifier state. The author reproduces this by routing the Jev endpoint to a local HTTP server and logging every request, showing that the user message drops out of a fixed 30-message window depending on the agent’s message shape, causing most later calls to be evaluated without the user’s explicit context. It also identifies a second issue—default risk criteria are never sent when criteria is left as None—so the request effectively doesn’t include the expected risk definition. Finally, it notes a fail-closed behavior where a single TypeSafe 429/503 classification error aborts the whole agent run, and a missing/incorrect “threshold” argument plus hard-coded 0.5 cutoff can cause additional failures. The fix is to pin the latest user message into Jev’s state for the whole run and to pass the criteria explicitly; the author provides a small subclass (“PinnedAutoMode”) and shows it eliminates the blind calls. The conclusion recommends using Auto Mode only for short tool-call chains and using the pinned approach (plus human-in-the-loop for destructive tools) for longer agent tasks.
Read the full blog for free on Medium.
Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.
Published via Towards AI
Towards AI Academy
We Build Enterprise-Grade AI. We'll Teach You to Master It Too.
15 engineers. 100,000+ students. Towards AI Academy teaches what actually survives production.
Start free — no commitment:
→ 6-Day Agentic AI Engineering Email Guide — one practical lesson per day
→ Agents Architecture Cheatsheet — 3 years of architecture decisions in 6 pages
Our courses:
→ AI Engineering Certification — 90+ lessons from project selection to deployed product. The most comprehensive practical LLM course out there.
→ Agent Engineering Course — Hands on with production agent architectures, memory, routing, and eval frameworks — built from real enterprise engagements.
→ AI for Work — Understand, evaluate, and apply AI for complex work tasks.
Note: Article content contains the views of the contributing authors and not Towards AI.
Related posts
Recent Posts
