I read that SAP led a round valuing n8n at $5.2 billion, and the number wasn’t what stuck with me. What stuck was the picture of everything you have to wire up before n8n moves a single byte for you.

n8n is good software. People love it for a reason, and the reason is real: you draw boxes, connect them, and events flow from one app to another while you sleep. But that whole graph sits on top of a layer most demos skip past quickly, which is the part where you go get credentials for every single app in the chain before any of it runs.

What “connect Gmail” actually means

So you want a workflow that reads new Gmail, drops rows into a Sheet, and updates a contact in your CRM. Three nodes. Looks like five minutes of work in the editor.

It is not five minutes.

Gmail wants an OAuth client. You’re in Google Cloud Console creating a project, enabling the Gmail API, configuring a consent screen, picking scopes, and copying a client ID and secret back into n8n. Sheets is another grant. Your CRM hands you an API key buried three settings pages deep, and if it’s HubSpot or Salesforce you’re now reading docs about which token type you even need. Each connection is a small bureaucratic errand, and you do all of them before you find out whether the workflow logic was right in the first place.

And every one of those grants is a long-lived key sitting in a server somewhere, scoped to your inbox, waiting to either expire at an inconvenient time or leak in someone else’s breach.

The thing the demo never shows you

Tokens rot. That’s the part nobody puts in the launch video.

A scope changes on Google’s side. A CRM rotates its API format. An OAuth grant hits its 7-day expiry because the app was still in testing mode. Suddenly the workflow that ran fine for a month throws a 401 at 3am and your “automation” is now a debugging session about refresh tokens. I’ve watched people spend more time maintaining the plumbing than they ever saved on the task the plumbing was supposed to handle.

This is the quiet cost of integration-first automation. You don’t pay it on day one. You pay it slowly, forever, every time something upstream shifts.

Your tab is already logged in

Here’s a different starting point. You already opened Gmail this morning. You’re already signed into your CRM in another tab. The browser is holding all of that authenticated state right now, and it cost you nothing extra because logging in is just a thing you already did.

A browser agent works inside that state. It doesn’t ask Google for a token, because it isn’t a third party knocking on Google’s door — it’s acting in the session you opened yourself, the same way a script you run in your own console would. No client ID. No consent screen. No key to rotate.

Dassi is built on exactly this idea. It’s a Chrome extension that lives in your side panel and operates the pages you’re already viewing, so connecting Gmail to a Sheet to your CRM isn’t an integration project. It’s three tabs you already had open. The agent reads the email, fills the spreadsheet, and updates the contact by clicking through the same interfaces you would, using the login you already have.

We wrote more about this shift in Browser Agents Are the New API, because the pattern shows up everywhere once you notice it. Most “no API” complaints are really “no API I want to spend a weekend authenticating against” complaints.

Where the two approaches actually diverge

n8n wins when you genuinely need machine-to-machine reliability. Webhooks firing thousands of times a day, server-side jobs that must run whether your laptop is open or not, pipelines other engineers depend on. That’s its home turf and a browser agent is a bad fit for it.

But most people automating Gmail and a spreadsheet and a CRM are not running a thousand jobs a day. They’re doing a fiddly thing a few times a week that’s too small to justify the OAuth setup and too annoying to keep doing by hand. For that whole middle zone, the integration tax is pure overhead, and it’s where the setup friction quietly eats every productivity gain the tool promised.

The model doing the reasoning can be identical. With Dassi you bring your own key, or just sign in with your existing ChatGPT or Claude subscription, so the intelligence is the same intelligence n8n would call. What’s different is the plumbing underneath it, and honestly the plumbing is most of the work.

A $5.2B bet on the hard part

SAP didn’t pay $5.2 billion for the AI. They paid for the connector library, the years of integration glue, the catalog of apps n8n already speaks to. That moat is real and it’s worth a lot.

It’s also a moat around a problem your browser doesn’t have. When the credential lives in your own session, there’s nothing to integrate. You can try Dassi from the Chrome Web Store and the setup is roughly: install it, open the tabs you were going to open anyway, and ask.

No client ID required. Which feels almost like cheating, until you remember the credential was always yours to begin with.