SAP put money into n8n last week at a $5.2 billion valuation, and my HN feed has been wall-to-wall with “Building N8n AI Automation Workflows” ever since. I get the appeal. You drag a few nodes onto a canvas, connect them with little wires, and suddenly your Gmail talks to your Notion which talks to your CRM. It looks like magic in the demo.

Then you try to build one.

The node that won’t connect

Every node in an n8n workflow is a service, and every service wants its own credential. Gmail needs an OAuth app you register in Google Cloud Console. Slack needs a bot token with the right scopes. Airtable wants a personal access token. Your CRM wants an API key buried three menus deep in some settings page. Before the workflow does one useful thing, you’ve spent an afternoon as a part-time identity administrator, clicking through consent screens and pasting secrets into form fields.

This is the integration layer, and it’s most of the actual work. The clever automation you wanted to build ends up sitting on a pile of plumbing that has nothing to do with your problem. I once counted forty connected tools in a setup that still couldn’t reliably fill out one web form, which is roughly the argument I made in 40 MCP Tools and It Still Can’t Fill Out a Form.

Your browser already did the login

I keep snagging on one thing while reading those tutorials. Every credential the workflow demands is a login I already have. I’m signed into Gmail. I’m signed into Slack. My CRM tab is open right now with a perfectly valid session cookie sitting inside it. The whole API-key dance exists to hand some server the access my browser was already given the moment I logged in this morning.

A browser agent skips that layer entirely. It doesn’t authenticate to Gmail’s API because it never touches Gmail’s API. It acts inside the Gmail tab you already have open, the same way you would, clicking and typing against a session that’s already yours. So there’s no OAuth app to register. No token to rotate when it silently expires next quarter. No scope you misconfigured that leaves the automation half-broken.

Dassi works exactly this way. It’s a Chrome extension that lives in your side panel and reads the page you’re looking at, so when you ask it to pull the last week of replies from a client and stitch them into a summary, it does that inside the tabs that are already logged in. Because the login already happened, the thing people spend a weekend wiring up in n8n is just… not there. There’s nothing to wire.

The credential you don’t create can’t leak

Every API key you mint for a workflow is a long-lived secret that now lives on somebody’s server, waiting patiently to be breached or misused. LiteLLM users found that out the hard way. The n8n model quietly turns your automation platform into a vault holding keys to your entire stack.

Two different bets

n8n’s whole design assumes the connection is the hard part worth building infrastructure around. And for a lot of backend work, it genuinely is. If you’re syncing a database into a warehouse every fifteen minutes while nobody’s awake, you want an API contract, retries, and a token that doesn’t depend on some tab staying open. A browser agent is the wrong tool for that, and I’m not going to pretend otherwise.

But a huge slice of what people actually build in these workflows is stuff they already do by hand in a browser all day. Read the email, find the relevant rows, draft the reply, update the record, slap a label on it. Those aren’t backend jobs. They’re the busywork of your own workday, and routing them through five registered API integrations is a strange amount of ceremony for tasks you could otherwise just do, in the tabs sitting in front of you.

I made this case before in a piece arguing that browser agents are basically the new API: when there’s no clean integration, or the integration costs you a weekend to stand up, acting through the same interface a human uses turns out to be faster than building the pipe. That post is Browser Agents Are the New API if you want the longer version.

There’s a BYOK angle here too. A browser agent holds none of your service credentials, and for the reasoning model you can bring your own key. Dassi lets you log in with your ChatGPT subscription or plug in Claude, Gemini, DeepSeek, whoever. The platform never becomes a credential vault, because there’s nothing to store.

Where it falls apart

I won’t oversell this. A browser agent needs the tab open and you roughly around. It won’t run at 3am while you sleep, and it won’t quietly grind through two hundred records with nobody watching. For always-on orchestration, n8n wins, and it isn’t close.

So which one

The trending tutorials all start from the same assumption, that automation begins with connecting services. Maybe that’s backwards for the kind of automation most people actually mean. The connecting was never the goal. It was the tax. And $5.2 billion is a lot of money riding on a tax that, for the browser-shaped half of the work, you already paid at login.

I’ll keep an n8n instance around for the boring cron stuff. But the moment a workflow is really just me, doing me-things, in tabs I’m already signed into, standing up six OAuth apps to automate it feels like renting a forklift to carry a coffee. If you want to try the browser-native version, Dassi is free and lives in your side panel.