SAP just put a $5.2 billion valuation on n8n, and the headline rocketed to the top of every tech feed I read this morning. That’s enormous money for a workflow automation tool, even by the standards of an enterprise software giant that buys things for sport. n8n is genuinely good at what it does, so I’m not surprised investors want in. What I keep circling back to is the quiet assumption buried inside the whole category, the one nobody puts on the pitch deck.

Workflow automation only works where there’s an API.

The part nobody prices in

n8n connects things. Slack to Notion, Stripe to a spreadsheet, a webhook to a database write. It does this beautifully, and when every system you touch exposes a clean API with stable auth, you can wire your whole company together and never click a button again.

But think about what your actual Tuesday looks like. You’re in a vendor portal that hasn’t shipped an API since 2019. You’re pulling numbers off an internal dashboard your finance team bolted together. You’re inside some government filing site, a supplier login, an old CRM, a booking system that exists only as a web page. None of these have a connector. Most never will, because the company behind them has no incentive to build one and no roadmap that includes you.

So the $5.2B bet is really a bet on the slice of work that lives behind an API. And that slice, in my experience, is maybe half. The other half happens in tabs.

Tabs are not a rounding error

Here’s where I think the automation conversation goes sideways. We talk like the API-less work is some shrinking legacy problem that’ll get cleaned up eventually. It won’t. The long tail of software is the software. There will always be more obscure web apps with login walls and no developer docs than there are clean REST endpoints, and a huge amount of real economic activity, the boring expense-report, copy-this-figure-into-that-form, check-the-status-of-an-order kind, lives precisely in that tail where n8n simply cannot reach without someone hand-building an integration that may never exist.

And when there’s no API, the workflow tool hits a wall. You can sometimes fake it with a scraper or a brittle HTTP node that breaks the moment the page markup shifts, but that’s not automation, that’s a maintenance contract you signed with yourself.

A browser agent doesn’t care whether an API exists. It reads the page the way you do, clicks what you’d click, types what you’d type. The website doesn’t need to expose anything, because the agent is already standing where you stand.

Whose login is it anyway

This is the second thing that gets glossed over. Workflow tools authenticate through OAuth tokens and stored credentials and connector configs. Somebody on your team grants n8n access to Salesforce, sets up the API key, manages the refresh, prays the scope doesn’t get revoked during an audit. That’s fine for ten core systems. It’s absurd for the hundred random portals you touch in a quarter.

A browser agent skips all of it. You’re already logged into that vendor site in your own Chrome session. The cookie is sitting right there. The agent uses the session you already have, so there’s no token to provision, no third-party app to authorize, no separate place your credentials get stored and possibly leaked. Dassi runs as a Chrome extension in your side panel and operates inside your existing tabs, which means the thing doing the clicking is authenticated exactly because you are. We wrote more about why cloud browser agents can’t see your tabs, and the login state is half the reason that distinction matters at all.

Two different shapes of automation

I don’t think browser agents replace n8n, and I’d be suspicious of anyone selling that line. They solve different shapes of problem.

n8n is for the durable, high-volume pipe between two systems that both expose APIs. Set it once, let it run a million times, sleep well. Worth $5.2B to SAP? Maybe. Enterprises run on exactly those pipes.

A browser agent is for the messy, occasional, human-shaped task where no pipe exists and building one would cost more than the task is worth. Pull the figures off this dashboard. Draft replies to these six emails. Fill this form using data from that other tab. It’s the work you’d do yourself in nine minutes, except now you don’t.

The half SAP didn’t buy

What strikes me about the timing is that all this enterprise money is flooding into automation right as the API-versus-browser line gets sharper, not blurrier. The connector model is winning the part of the market it was always going to win. The rest, the tabs, the logins you already have, the sites that’ll never ship an SDK, is still wide open and barely contested.

If you want to see what the other half feels like, install Dassi from the Chrome Web Store and point it at whatever portal has been wasting your afternoons. Bring your own model key, or just log in with the ChatGPT subscription you’re already paying for.

SAP wrote a check for the systems that talk to each other. Nobody’s written one yet for the systems that only talk to you.