n8n is not being disrupted out of existence. It is being disrupted out of the build step. Drawing the workflow was the product, and an external coding agent now does that faster and for free. What survives is the runtime: credentials, retries, scheduling, replay, audit, and the operational burden nobody wants to own. That is a real business and a durable one, and it is currently priced per execution, which is exactly backwards for a world where an agent generates the executions.
The question people keep asking is whether visual workflow automation is dead now that a coding agent can write, test and deploy an integration in one sitting. That question is pitched at the wrong altitude, so it produces an answer that is either obviously right or obviously wrong depending on who is arguing. Automation platforms sell two separable things: a place to author the work, and a place to run it. Authoring just got commoditised by something the vendor does not own and cannot price. Running did not. n8n's own documentation now tells customers to describe the workflow they want in their favourite AI tool, which is the clearest concession available that the canvas is no longer where the work happens. What is left is credentials under rotation, replayable runs, approval gates, retention and somebody accountable at three in the morning. That is a durable business. It is also smaller and more price-exposed than a 5.2 billion dollar valuation implies, and it is currently metered per execution, which is exactly backwards when an agent is the thing generating the executions.
Ten findings in the report, each sourced and labeled by evidence type. Six of them carry the argument.
n8n ships an instance-level MCP server that lets Claude, ChatGPT, Claude Code, Codex, Cursor, Windsurf and Gemini CLI build, edit and run workflows inside a customer's instance from outside it. The documentation frames this as describing the workflow you want in your favourite AI tool. That is a concession, and it is the healthiest thing in the company's recent product history.
At published list prices, Starter works out at roughly 0.80 euro cents per workflow run, Pro at 0.50, and Business at 1.67. Business costs 3.3 times as much per run as the tier below it. A normal volume ladder gets cheaper as you climb it. This one is a governance ladder wearing a throughput ladder's clothes, and everyone involved knows it.
Ten thousand workflow runs cost 50 euros on n8n Pro. Ten thousand ten-step workflows on Cloudflare Workflows fall inside the 500,000 steps included in a 5 dollar Workers Paid plan, with marginal steps at 0.80 dollars per hundred thousand. n8n sells the catalogue, the credential store and the operations, which is a fair thing to sell. It is not a fair thing to price by the run.
GitClear's analysis of 623 million changes from 2023 to 2026 found code block duplication up 81 percent, within-commit copy and paste up 41 percent, error-masking constructs up 47 percent, and refactoring line moves down 70 percent. Changes touching code older than twelve months fell 74 percent. The write step got cheap. The keep step did not, and it is now being skipped.
OpenAI told developers on 4 June 2026 that Agent Builder, launched at DevDay in late 2025, would be unavailable from 30 November 2026, pointing them to the Agents SDK for code and to Workspace Agents for prompting. A canvas backed by the largest balance sheet in the category lasted about thirteen months. Migration discussion in the thread settled on Lambda and Cloudflare Workers.
Self-hosted community users generate no telemetry n8n monetises and no cancellation event when they stop building. A heavy user who still runs every workflow they built, and has quietly stopped building new ones, looks identical in the data to a healthy one. If the loss is happening at the authoring step, that is precisely where it hides.
Treating workflow automation as a single category is what makes the disruption argument unresolvable. Split it and each piece has a clear answer, and they point in different directions.
A revenue operations analyst who needs a Slack alert when a deal stage changes is not opening a terminal, and the fact that an agent could write that integration in ninety seconds does not help her, because she cannot review it, deploy it or fix it at 2am. Zapier has owned this for a decade.
n8n competes here awkwardly · assumes JSON literacy
A backend engineer used the canvas because wiring six APIs together by hand was tedious. The agent now finds the tedium cheaper than opening the canvas. This buyer was never paying for the visual metaphor, they were paying to skip boilerplate, and boilerplate is exactly what got commoditised.
New-build volume decays first · old workflows keep running
An organisation that has to answer for what ran wants credentials under rotation, a failed run replayable with its exact input, an approval step, and a list an auditor can be shown. No coding agent addresses this, and every coding agent makes it more urgent, because the volume of things running unattended just went up.
Served worst by per-execution pricing
The full twenty-five pages carry the pricing arithmetic, the competitive table naming who actually threatens the position, four scenarios to mid-2028 with observable tripwires, the leading-indicator table anyone can watch from outside the company, and the one measurement n8n could run today that is worth more than every market forecast quoted here.
This isn't a vendor summary. Every sentence is labeled by what stands behind it: verified fact, vendor claim, third-party estimate, my assessment, hypothesis, or scenario. Sources are numbered and clickable. Forward-looking sections use scenarios with observable tripwires, not forecasts. It's the same method behind every market assessment I write.
Twenty-five pages, built from public sources with no client brief and no interviews. Read it in the browser or take the PDF.
Each report here answers a real question, directed and researched against public sources and evaluated against a stated assumption, then delivered as Word and PDF. If you're weighing a platform, sizing a category, or defending a number to a board, tell me the decision behind it and I'll tell you honestly whether a report is the right tool.
Commission an assessment