SAP invested in n8n today at a $5.2 billion valuation, and the number is doing what a number like that always does, which is making half my feed argue that automation is the biggest category in software and the other half argue that it’s 2021 all over again. I think the demand is real. I also think the demo videos skip the boring part.

The boring part is the credential.

The auth layer nobody screenshots

Before a single node fires, somebody creates an OAuth app in a Google Cloud project, picks scopes, waits on a consent screen, pastes a client secret into a credential vault, and then does the whole ritual again for Slack, again for Notion, again for whatever CRM sales picked last quarter. None of that is in the tutorial thumbnail. The thumbnail is the canvas with the pretty boxes.

I don’t think this is n8n’s fault. It’s the price of the model. If your automation runs on a server with no human attached, the server has to prove who it is, every time, to every service. Tokens are how it proves that.

400 connectors and still no connector

There was a Show HN a while back where someone built a browser agent specifically because Reddit’s API wasn’t officially available to them. Not expensive. Not rate-limited. Just closed. That’s the case that breaks the connector model entirely, and it’s more common than the integration-count marketing suggests.

Count the tools your team actually touches in a week. Some of them are Gmail and Slack and Sheets, and those have connectors, obviously. But then there’s the regional payroll portal, the freight broker’s tracking page, the university library system, the county permit lookup, the internal admin panel that a contractor built in 2019 and nobody has redeployed since. The insurance carrier portal your ops person has used for nine years, where nobody on the vendor side has ever said the word “webhook” out loud. That’s the long tail, and it is where a shocking amount of actual work lives.

For any of those, the n8n answer is: write a custom node, or wire up an HTTP Request node and reverse-engineer whatever the site’s frontend is calling, and maintain that forever, including the day they rotate the session cookie format. Doable. Genuinely a lot of engineering for what a person does in four clicks.

So the connector count is a little misleading as a measure of coverage. It measures how many of the top-N SaaS products have public APIs and a maintainer who cared, which correlates weakly with the tools any specific company depends on. I wrote more about that gap in Browser Agents Are the New API.

Your tab already did the auth

Here’s the shift that took me a while to appreciate. You are already authenticated. Right now, in Chrome, there’s a session cookie for every one of those systems, established through whatever SSO and MFA dance your IT department mandates, and it’s sitting there doing nothing between clicks.

A browser agent runs inside that. Dassi lives in the Chrome side panel, sees the page you see, and clicks the things you’d click. There’s no OAuth app to register, because there’s no third party asking for access. The permission model is your existing login.

“No official API” stops being a blocker. So does “no connector exists.”

Where n8n wins, and it’s not close

Anything scheduled and headless. A pipeline that runs at 3am, processes eleven thousand records, retries on failure, logs to your observability stack, and doesn’t need a laptop to be awake. Browser agents are bad at that. They’re slower per operation, they need a browser running, and they’re not the thing you want on a cron for a high-volume ETL job.

Two different shapes of work

I’d put it this way: n8n is for the workflows you’ll run ten thousand times, and a browser agent is for the ones you’ll run eleven times against a system nobody will ever build a connector for.

The eleven-times category is enormous and mostly invisible, because it never gets automated at all. It just gets done by a human on a Thursday afternoon, tabbing between a portal and a spreadsheet. Nobody writes an integration for that, because the integration costs more than the task. But the marginal cost of pointing an agent already sitting in your side panel at a page is roughly zero, which changes what’s worth automating in the first place.

Same thing happened with Reddit’s pricing, which I got into here. The API had a price. The browser didn’t.

SAP’s check says the automation demand is real. I’d just note that the demand and the connector list have never really matched up, and $5.2 billion doesn’t fix that. Somebody still has to go make the OAuth app.