n8n launched Assistant today, September 9, and if you've ever spent an afternoon wiring up a workflow that should have taken twenty minutes, this one matters. The pitch is simple: describe the automation you want in a sentence, and Assistant plans it, builds it on your canvas, runs it, reads the errors, fixes them, and hands you a working workflow. The reality is more nuanced, but the direction is worth paying attention to.
The gap it's trying to close
If you've used n8n — or Make, or Zapier, or any automation tool — you know the real bottleneck isn't the platform. It's the translation step. You know what you want: "when someone fills out my contact form, enrich the company, check if they're already in my CRM, and either route to sales or send a templated reply." Turning that sentence into eight correctly configured nodes with the right credentials, filters, and error handling is where the afternoon disappears.
n8n Assistant targets that gap directly. You type the outcome, it asks clarifying questions when your request is ambiguous ("which form? which channel?"), and then it builds the workflow on the same canvas you'd use by hand. No separate product surface, no hidden configuration, no magic. The nodes appear, wired together, and you can open each one and see exactly what it's doing.
What actually happens when you use it
The flow goes like this: you describe what you want, Assistant asks follow-up questions if needed, it builds the nodes, it asks for credentials when it reaches a node that needs an account connected (no upfront setup checklist), and then it runs the workflow against a test execution. If something fails — say the enrichment API returns an empty array for sole traders — Assistant reads the execution data, figures out what went wrong, proposes a fix, applies it, and runs it again.
That debugging loop is the part that separates this from "generate a workflow from a prompt." Plenty of tools can do that. The question has always been what happens at 2 AM when a node breaks and nobody remembers how it was supposed to work.
The workflows Assistant builds are standard n8n workflows. They live in your project, they show up in your execution history, they're editable by hand, and they're owned by your team — not by whoever wrote the prompt. A week later when the routing rules change, someone who never touched the original prompt can open the canvas, see every node, and adjust the filter. That's the part that matters for anyone running automation in a business context.
Why this matters beyond n8n
The broader trend here is that automation platforms are becoming AI platforms. We covered this shift earlier this year when Zapier changed its pricing to charge per model tier, and when Make started integrating LLM nodes directly into scenarios. n8n has been aggressive on this front too — native LangChain nodes, an AI Agent node that can call tools and route between models, and now Assistant.
But there's a meaningful difference between "your automation tool has AI features" and "your automation tool builds automations for you." Assistant is the second thing. And the design choices n8n made — build on the existing canvas, keep workflows inspectable, require confirmation for credentials and activation — suggest they understand the failure mode. The problem with fully autonomous AI agents isn't that they can't do the work. It's that nobody can explain what they did afterward.
This is the same tension showing up everywhere in the AI tooling space. Coding agents can write code, but can your team review it? Customer support bots can answer tickets, but can you audit the logic? The tools that survive long-term in production are the ones where the output is something a human can inspect, modify, and hand off. n8n Assistant gets that, at least in its design.
What to watch for
The real test is what happens in the first month. Can Assistant handle multi-step workflows with branching logic and error handling? How does it perform when credentials expire mid-run? Does it play well with custom code nodes, or does it try to replace everything with built-in modules?
If you're already running n8n self-hosted, Assistant is worth trying today. Describe a workflow you've been putting off — lead routing, invoice processing, Slack notifications with conditional logic — and see what it builds. You'll know within ten minutes whether it saves you time or just produces something you need to rewrite anyway.
If you're on n8n.cloud, it should be rolling out now. The workflows it creates don't require any special version or configuration — they're just n8n workflows, which means they'll keep running even if you never touch Assistant again.
The automation space has been talking about "describe what you want and it happens" for years. Today is the first time it's felt close to real, at least for the kind of workflows most small businesses actually need.