The number that stopped me was $5.2 billion. That’s what SAP just paid to back n8n, the workflow-automation tool that until pretty recently was a German open-source side project people ran on a Raspberry Pi. When the most boring, profitable enterprise software company on earth writes a check that size, it’s not buying a tool. It’s buying a thesis: automation is the prize everyone’s fighting over now.

I agree with the thesis. I just think n8n is the slow way there.

What you actually do when you build an n8n workflow

Say you want something simple. When a new lead fills out a form, look them up, draft a personalized reply, and log it in your CRM. In n8n you open the canvas and start dragging nodes.

First the trigger node. Then an HTTP node, or better, the dedicated CRM node if one exists. That node needs credentials, so you go create an OAuth app in the CRM’s developer console, copy the client ID and secret, paste them back, run the OAuth dance, hope the scopes are right. Now the enrichment service. Another node, another API key, another rate limit to read about. Then the email node, which wants its own auth. Then a Set node to map fields because the CRM calls it firstName and the form calls it first_name, and these things never agree.

By the time it works you’ve touched four developer consoles and stored five secrets, and you’ve built a small distributed system whose only job was to do something you could’ve done by hand in ninety seconds. The workflow is genuinely powerful once it runs. Getting it to run is the tax.

The plumbing is the product

Here’s what I keep noticing about every integration-based automation platform, n8n included: most of the engineering isn’t the automation. It’s the connective tissue. The nodes, the credential vault, the field mappers, the retry logic for when an API returns a 429. n8n is good at this stuff, which is exactly why SAP wanted it. Enterprises drown in that tissue.

But all of that machinery exists to solve one problem, which is that the automation engine lives somewhere your accounts don’t. It runs on a server in a data center, and that server has never logged into your CRM, so every single service has to be re-introduced to it through a connector and a credential. The whole architecture is a workaround for the fact that the robot isn’t you.

Your browser already passed the auth check

Now think about where the same task lives from your seat. You’re in Chrome. You’re logged into the CRM, into Gmail, into the enrichment tool, all in tabs, all with live sessions your browser is holding open right now. The hard part that n8n spends nodes and OAuth apps reconstructing? Your browser solved it this morning when you typed your password.

A browser agent runs the automation right there, inside those already-authenticated tabs. Dassi is a Chrome extension that sits in the side panel, sees the page you see, and acts on it. So the lead-reply workflow becomes a sentence: read this form submission, look the person up in the tab I’ve got open, draft a reply in Gmail, and add a note in the CRM record. No node for the CRM, because the CRM is already on screen. No OAuth app for Gmail, because you’re already in Gmail. No field mapping, because the agent reads the actual page the way you do, labels and all. There’s nothing to connect because nothing was ever disconnected. The credential problem that defines half of n8n’s surface area just doesn’t appear, and that absence is the entire point.

Where this breaks down, honestly

I’m not going to pretend the browser approach wins everywhere. n8n shines when something needs to run at 3am with no one watching, ten thousand times, forever. Server-side workflows are built for unattended scale, and a browser agent in your side panel is not that. It runs when you’re there, on the tasks in front of you.

But that describes a huge amount of real work. The lead reply. The weekly report pulled from three dashboards. The fifty rows you’re copying between an admin panel and a spreadsheet because the export costs $100 more a month. None of that needs a data center. It needs something that can already see your screen.

The bet under the bet

SAP’s $5.2 billion says automation is where the value is. Fine. I’d take that bet too. The part I’d quibble with is the assumption that automation has to mean building integrations, that the path to a robot doing your work runs through a graveyard of OAuth apps and API nodes you maintain forever.

There’s a second path, and it’s the one browser agents took while everyone was drawing nodes on a canvas. Don’t connect the agent to your services. Put it where your services already are. You can grab Dassi from the Chrome Web Store and point it at whatever’s in your tab right now, no client secret required.

Whether that’s worth $5.2 billion, I have no idea. n8n’s accountants seem confident, though.