Relay is shutting down and its staff are joining Google’s Chrome team. I read that headline twice, not because acqui-hires are interesting, but because of where they landed.

Relay built the thing everyone in this category builds: a cloud platform where you assemble AI steps into workflows that reach your SaaS apps through connectors. Good product. Funded well enough to hire people Google wanted to keep. And the landing spot wasn’t a workflow company, or an agent-infrastructure company, or some orchestration layer startup. It was the browser team.

The connector tax nobody itemizes

Every integrations backend opens with the same request. Authorize us. OAuth into Gmail, mint an API key for Notion, spin up a service account for the CRM, then wait four days for someone in IT to sign off on the Salesforce scope you asked for. Repeat per tool, forever.

Every run of a connector-based workflow begins the same way, with a machine that has never met you presenting a token it was handed weeks ago, hoping the permissions still hold and that nobody rotated anything over the weekend.

That’s the model SAP just valued n8n at $5.2 billion for, and it deserves the valuation, because for the roughly twenty percent of your stack with clean public APIs it is genuinely excellent. The other eighty percent is where the model falls over. Internal dashboards. Vendor portals. Your payroll provider’s admin panel. A state licensing site built in 2009 with no API, no developer docs, and no OAuth app you could register even if you wanted one. Just a login form and a session cookie.

Chrome already knows who you are

Your browser is the largest pile of live authorization you own, and nobody treats it that way. Forty-something tabs, half of them behind SSO, sessions your company’s IdP already blessed this morning. No token to provision. No scope review. The permission grant happened when you logged in, the way it always does.

An agent that runs inside that tab inherits all of it. It doesn’t authenticate as a service principal with a name like relay-prod-worker-3. It acts as you, in your session, with exactly the access you have and not one scope more. When your access is revoked on Friday, so is the agent’s, automatically, because there was never a separate identity to go clean up.

This is the whole design premise behind Dassi. It lives in the Chrome side panel and works in the tab you already have open. Draft a reply in the Gmail thread you’re staring at. Pull the line items off an invoice in a vendor portal that will never ship an API. Fill the same intake form forty times without anybody wiring up a connector, because there is no connector, because the site is the connector.

The cloud browser agents shipping on Show HN every week get this half right. They understand the browser is the substrate. But their browser runs in a data center, starts blank, and hits a login wall on the first navigation, which is a problem we’ve written about before and which no amount of proxy rotation fixes.

You can watch it

Which is the underrated part. The agent moves, you see it move, you stop it when it goes somewhere dumb. Cloud automation fails silently at 3am and you find out from a customer.

Where this gets uncomfortable

The browser-native position has real costs and I’d rather name them than pretend. Your machine has to be awake. Nothing runs at 3am unless you leave Chrome open, which is a hell of a downgrade from a workflow engine that hums along in a Frankfurt data center while you sleep. Scheduled jobs, retries, fan-out across ten thousand records — connector platforms win all of that, decisively.

So the honest split is: n8n and its descendants own the machine-to-machine work, the nightly syncs, the webhook plumbing. Browser agents own the work that only exists because a human sits in front of a screen and clicks. That second category is enormous and mostly unautomated, and it has been sitting there the entire time, unclaimed, because everyone kept building for the first one.

More on why the integration layer keeps missing it.

I don’t know what Relay’s team will actually build inside Chrome, and the announcement said nothing useful about it. Maybe nothing ships for two years. But a group of people who spent years building cloud automation looked at the map and walked toward the address bar, and that movement is worth more than any roadmap they could have published.

Google now owns the browser and the people who learned the hard way that automation belongs in it. Somebody at n8n should probably be watching.