Agentic Layer Check
n8n's meter counts one thing: the number of times a workflow runs, whatever's inside it. That was a sensible unit for connecting a webhook to a Slack message. It's a stranger unit once one of the nodes in between is an AI agent that can, on its own, decide to do thirty times the work. Here's how the pricing reads against the nine layers.
What they sell
A workflow automation platform, cloud hosted or self hosted, for connecting apps and building automations through a visual, node based editor: increasingly marketed around building AI agents, not just syncing SaaS tools.
Who they sell to
Technical and ops builders working self serve from Starter through Business, plus platform and IT teams standardising automation and agent building across a wider organisation on Business or custom Enterprise.
How they price
Four tiers metered on monthly workflow executions "regardless of complexity": Starter (€20/mo, 2.5K executions), Pro (€50/mo, 10K executions), Business (€667/mo, 40K executions, self-hosted), and custom Enterprise, each with its own concurrency cap and a separate, non-topupable AI Assistant credit allowance.
Two choices here are genuinely worth crediting, not just noting on the way to the critique.
Unlimited users, unlimited workflows and unlimited integrations ship on every paid plan, Starter included. n8n's own pricing page draws the line at exactly one thing: monthly executions. That's a genuinely uncommon choice in a category that usually taxes headcount and connector count on top of usage, and it means the meter tracks something closer to what a team actually produces than to how many people happen to hold a login.
Because there's no per seat cost, any team member can open the editor, wire up a workflow, and ship it without a separate purchasing conversation first. That's a real adoption advantage: the fastest way to build habitual use is to remove the friction between wanting to build something and being allowed to, and n8n removes more of that friction than most tools in this category.
The unit that made sense for connecting apps stops making sense once one of the things it connects can decide, on its own, to do far more work.
n8n's own pricing page states it plainly: cost doesn't depend on "how many steps are in the workflow or how much data it processes," and its own documentation confirms an execution is a single run of a workflow, however many nodes it contains. That's a reasonable place to put the meter for deterministic automation, where a two node webhook and a fifty node data sync cost roughly the same to run. It stops being reasonable once one of those nodes is an AI agent: n8n's own documentation also confirms that calling a workflow through the Execute Sub-workflow node counts only the parent execution, so a multi agent system built by chaining ten agent sub-workflows together still bills as one execution, regardless of how many LLM calls happen inside it. The unit built for "how many things did you connect" doesn't hold once one of the things connected can, by itself, do thirty times the work for the same one line on the invoice.
AI Assistant credits sit on a separate allowance from executions (2,300 a month on Starter, up to 13,700 a month on Pro, per n8n's pricing page) and, per n8n's own help center, "reset on the 1st day of each month" and "cannot be manually topped up, carried over, or reset outside of this monthly schedule." A customer who wants to keep using the assistant mid cycle has no self serve way to buy more: the only lever is to wait for the calendar to turn over, or upgrade an entire plan tier for headroom on a completely different meter. That's a Commercial Systems gap rather than a pricing decision: the billing stack has no top up mechanism for the exact moment a customer is asking to spend more.
The flat per execution unit doesn't stay contained to Layer 6. It lands two layers over on Revenue Architecture (Layer 8): the infrastructure cost of a two node webhook and a forty step, thirty LLM call agent aren't remotely close, but both draw down the same single unit of the same monthly allotment, so a workspace shifting its workload toward heavier agents sees its true cost to serve climb well before the number on its invoice moves at all — it only moves once a customer crosses an execution ceiling entirely. It also lands on Sales Strategy & Enablement (Layer 4): a rep pitching n8n's growing library of AI agent templates has to explain why a customer's most complex, most valuable workflow costs precisely the same as their simplest one, which cuts against the value story the AI agent pitch is trying to tell in the first place. Neither cost shows up on the pricing page; both show up later, in gross margin and in the sales conversation, which is the pattern this series keeps finding: the layer where a decision gets made is rarely the layer where its cost lands.
Part of the Agentic Layer Check series: named companies, read against the nine layers. Want your own read? Request a teardown →
This is a public signal read: n8n's own pricing page, its own execution and sub-workflow documentation, and its own help center policy on AI credits, not a private briefing or company input. It reflects what's publicly documented as of the date above: pricing pages and credit terms change, and this piece will note it if n8n's does. n8n was not consulted on or paid for inclusion; see the disclosure approach this work is run under.
Sources: n8n: pricing → · n8n docs: understand executions → · n8n docs: break workflows into smaller parts → · n8n help center: AI credits FAQ →