n8n 2.39.0 Hardens AI Agent Workflow Tools

n8n shipped version 2.39.0 on September 8, 2026, and the change worth your attention isn't a new node. It's a fix for the exact failure mode that trips up production AI agents: workflow tools that go missing, get unpublished, or drift out of compatibility mid-run. If you're running n8n's AI Agent node with workflow-based tools, this is worth three minutes before your next deploy.

What changed

n8n 2.39.0 landed with more than 60 fixes, but three stand out for anyone building agents:

• Agent workflow tools are now hardened against unpublished, missing, or incompatible workflows. The release notes name this protection explicitly, which tells you n8n has seen this fail in the wild often enough to fix it directly.
• Agents track waiting workflow tools as background jobs instead of just blocking until a call resolves.
• Publishing an agent now rejects the action and names the specific unpublished workflow tools that would break it.

The release also touches the Anthropic Chat Model and AI Agent nodes, which now send disabled-thinking signals to the provider and parse content-block output correctly, and the AI Assistant picked up capability awareness for agent modifications. Outside the agent work, 2.39.0 adds Microsoft Teams online-meeting and channel-message operations, Atlassian Service Account (2LO) authentication for Confluence and Jira, and an OrgUnitPath option for G Suite Admin.

Full release notes:

github.com/n8n-io/n8n/releases/tag/n8n%402.39.0

Why it matters

Anyone who has built a multi-workflow agent in n8n knows the pattern: break a big agent into smaller sub-workflows, wire each one up as a tool, call it done. It works fine until someone edits or unpublishes one of those sub-workflows without realizing an agent depends on it. Before 2.39.0, that failure surfaced at runtime. The agent would try to call a tool that no longer existed or no longer matched its expected shape, and you'd hear about it from a support ticket, not a build error.

The new hardening catches this at publish time instead. Try to publish an agent that depends on an unpublished workflow tool, and n8n now tells you which tool is the problem before the agent goes live. That's the difference between a broken agent shipping quietly and a build failing loudly, with a name attached to the cause.

Background-job tracking for waiting workflow tools helps for a related reason. Agents that call slow sub-workflows, anything hitting an external API or waiting on a human-in-the-loop step, used to just sit there hoping the call resolved. n8n now treats that wait as a trackable job, so there's somewhere to look when an agent seems stuck instead of just seeming dead.

What it looks like in practice

Say you have an AI Agent node routing customer support requests to three sub-workflow tools: one that looks up order status, one that checks refund eligibility, one that escalates to a human. Before this release, if a teammate unpublished the refund-eligibility workflow to fix a bug, the agent kept running. Any customer who triggered a refund question got a silent failure or a generic error, and you found out after the fact.

With 2.39.0, republishing that same agent fails immediately, naming "refund-eligibility-check" as the missing tool. You fix it before the agent goes back live, not after a customer complains about a refund that went nowhere.

What to do about it

• Update to n8n 2.39.0 or later before your next agent deployment, and republish existing agents that use workflow-tool patterns to trigger the new validation.
• Audit any agent-as-tool setup you already have in production. This is a good moment to confirm which sub-workflows are load-bearing for which agents.
• If you use the Anthropic Chat Model node with extended thinking toggled off, check that downstream logic still parses responses correctly under the new content-block handling.
• Keep an eye on the next few releases. n8n has been iterating on agent tooling fast this month, and this probably isn't the last fix in this area.

The short version

Unpublished or incompatible workflow tools now fail loud, at publish time, instead of failing silent, at runtime. If you're running production agents built from workflow tools in n8n, update before your next deploy and republish to catch anything broken.