Google pushed a Workspace update today and pitched its new AI as your “office intern.” And the promise is that the intern reads Gmail threads, drafts replies, skims Docs, pulls numbers out of Sheets, and generally keeps things moving so you can focus on actual work.

But an intern who only works inside Google Apps is not really an office intern. That’s a Google Apps intern, which is a job title that does not exist outside of Mountain View.

The thirty-tab day you already have

Pull up Chrome on any knowledge worker’s machine. You’ll find something like this open: two Gmail tabs, a Google Doc, a handful of Linear tickets, one or two Notion pages, a Salesforce record mid-edit, a GitHub PR, a Stripe dashboard, Figma, Slack, and a Zendesk queue that has never once dropped below fifty. And the browser remembers every login. Every session cookie knows exactly who you are on every one of those services.

But none of that is visible to the Workspace intern. It sees the Google silo and politely nothing else. Ask it to “pull the revenue number from Stripe and paste it into this quarterly Doc” and it will nail the second half and completely skip the first.

The intern, expanded

It reads email. It writes Docs. It cannot log in to anything.

The silo is not a bug

Google didn’t forget about Notion. The intern ignoring Salesforce won’t be patched in a later update. But the whole product surface is scoped to what Google already has the rights to read, index, and train on. Asking a Workspace AI to cross over into Linear is asking Google to become the default router of your company’s entire SaaS stack, and that deal is not coming bundled into a thirty-dollar-per-seat plan.

So the silo is the business model. Not a missing feature.

What a real intern would do

Human interns do not request API access. When you ask a human intern to hunt down a vendor invoice, they click through Bill.com in the browser you handed them. And they peek at your calendar. They scroll Slack. They climb into that one ancient internal tool that nobody has documented since 2019. Because they use whatever tabs were already signed in on the laptop you gave them. The intern doesn’t call IT to ask for read access to the Zendesk admin panel. The intern uses the Zendesk admin panel that is already open on tab fourteen.

But browser agents match that pattern almost exactly. Dassi sits in the Chrome side panel and reads the page you’re reading. If you’re logged in, it’s logged in. No OAuth ceremony per SaaS app. No integration catalog. No pricing tier deciding which of your own tools the agent is “allowed” to touch. And crucially no pause where you explain to the agent that yes, your company really does use six different project trackers, and no, you will not be consolidating them this quarter, thank you.

The integration arms race is not winnable

Every AI vendor has a page bragging about their integration count. Forty connectors. Eighty. And the list grows every quarter, and the list still does not include that one damn internal dashboard your team shipped on Retool six months ago. Or the expense tool finance switched to last week. Or that industry forum you monitor that has no API at all.

Browser agents step around this whole race. They don’t need connectors. They need a rendered DOM and a logged-in cookie. We’ve written about this in the case against cloud browser agents that can’t see your tabs and in the post on ChatGPT’s app integrations still missing the page you’re on, so I’ll spare the recap.

But what’s interesting is that “intern” is an unusually honest metaphor for what Google shipped, even if the marketing team didn’t mean it the way I’m reading it. An intern who refuses to leave the Google building is not going to get promoted.

A small test, done before lunch

I tried the Workspace intern on a real task this morning: pulling Zendesk ticket volumes into a quarterly slide deck. It built the deck scaffolding. But it declined to go fetch the data. So I opened Dassi in the side panel, pointed it at the Zendesk view I already had loaded, and it filled the numbers in while my coffee was still hot. Hardly a scientific benchmark. But the unfairness is the entire point — one of them could see the open tab and the other one structurally cannot.

Where you’re left

Workspace AI is fine for the workday that lives entirely in Gmail, Docs, and Sheets. But for anyone whose day crosses six logins before noon, the intern needs to see the browser, not one vendor’s walled garden. You can grab Dassi from the Chrome Web Store and point it at whatever model you like — GPT-5.2, Claude, Gemini, the DeepSeek model you’ve been meaning to try.

Google’s intern will get better. But it will not, I suspect, ever learn to log into Salesforce on your behalf. Some walls are load-bearing.