SAP led a round that values n8n at $5.2 billion, and the tech press more or less handed node-based automation the crown for the quarter. I read the headline twice. A German enterprise giant writing that size of check at n8n is a real signal, and I don’t think the valuation is crazy given how many teams have quietly standardized on it. The thing is, the news got framed as “AI automation just won,” when what actually won is a tool that still makes you assemble the automation by hand before a single thing runs.

That gap is the whole story, and almost nobody covering the funding bothered to mention it.

What five billion dollars of workflow tooling still asks of you

Open n8n and you get a canvas. Drag a trigger node, drag an HTTP node, drag a transform, draw the lines between them. It’s elegant. It’s also a build step, and the build step is the work.

Before your “automation” automates anything, you have to know which API endpoint to hit, what shape the payload is, where the rate limits sit, and which of the 400-odd integrations maps to the SaaS tool you actually use. n8n didn’t invent that tax. Zapier, Make, Workato, all the node-and-connector tools live in the same world. You are still the integration engineer. The AI parts that got bolted on over the last year help you write a transform faster, but they don’t remove the canvas. The canvas is the product.

The OAuth screen they leave out of the launch video

Here’s where it gets less fun than the demo: credentials.

Every node that touches a real service needs auth. So you go generate an API key, or you walk through an OAuth consent flow, or you set up a service account and paste a JSON blob into a credentials manager. Then you do it again for the next service. Then a token expires three weeks later and your “set it and forget it” pipeline silently stops, and you find out because something downstream is empty.

I’ve watched people spend an entire afternoon getting a single Google Sheets-to-Slack flow authenticated, debugging scopes, re-authorizing, copying client secrets between tabs. The flow itself was four nodes. The plumbing was the whole day. That’s not a knock on n8n specifically. It’s the cost of any system that sits outside your session and has to be granted its own separate keys to everything you already have access to.

And those keys are real surface area. You’re handing a third-party automation platform standing permission to your inboxes, your CRM, your file storage. When something leaks, it doesn’t leak one tab, it leaks the token.

A different starting point

What if the automation just used the tabs you already have open and logged into? No nodes. No connector for a tool that doesn’t have a public API. No second copy of your credentials sitting in someone’s vault.

Watching what you do beats describing what you want

A browser agent comes at this from the opposite direction. Instead of you describing a workflow as a graph of API calls, it works the way you already work: in the browser, in the pages you’re authenticated to, clicking the things you’d click. That distinction sounds small until you’ve done both, and then it feels like the difference between writing a recipe and just cooking dinner.

Dassi lives in a Chrome side panel and acts inside your real session, which means when it goes to draft a reply in Gmail or pull line items out of a vendor portal, it’s already logged in as you, with no OAuth handshake and no API key to provision, because it’s literally using the tab you opened. There’s no integration to wait for either. If a tool renders in a browser, the agent can operate it, even the internal SaaS dashboards and government portals and half-broken admin panels that will never ship a public API in their lifetime. We wrote more about that idea in Browser Agents Are the New API, and it holds up better the more obscure the tool gets.

The credential story flips too. Your data and your logins stay in your browser. Dassi is free, runs on your own LLM key or your existing ChatGPT login, and doesn’t park a copy of your tokens on someone else’s server. You can grab it from the Chrome Web Store and have it doing real work in the time it’d take to authenticate one n8n node.

When the canvas is actually right

I’m not going to pretend node graphs are obsolete. If you need a thing that fires at 3am with no human anywhere near it, runs server-side, retries on failure, and processes ten thousand records on a schedule, a workflow engine is the correct tool and a browser agent is not. That’s a genuine job and n8n is very good at it. SAP didn’t put $5.2B into a toy.

But most of what people call “automation” isn’t that. It’s the forty browser tasks a week that are too fiddly to be worth building a pipeline for and too repetitive to keep doing by hand. Setup friction is exactly why those tasks never get automated, a point we kept running into in Setup Friction Is Why AI Productivity Gains Are Stuck. The canvas asks you to invest before you save. The agent just starts.

So the funding is a milestone for node-based tooling, and good for the n8n team, genuinely. It also quietly proves the limit of the model. Five billion dollars, and you’re still the one drawing the lines between the boxes.