Asana Bought StackAI. Your Agent Still Can't See the Tab You're Logged Into.
Asana announced it’s acquiring StackAI, the no-code platform that lets you drag and drop your way to a working AI agent. I caught it in my feed somewhere between two other “build your own agent” launches the same week, which tells you roughly where the whole industry’s head is at right now.
Every company wants to be the place you build your agent. Asana now. Salesforce with Agentforce. Microsoft with Copilot Studio. The pitch barely changes between them: describe what you want, click together a few connectors, and a tireless little digital worker takes it from there.
I actually like the ambition behind this. But there’s a gap sitting right in the middle of the promise, and nobody on the acquisition side seems eager to bring it up.
The agent runs somewhere you aren’t
StackAI agents live in the cloud. That’s not a detail, that’s the entire product. You assemble a flow, it executes on their infrastructure, and it reaches your other tools through API connectors and OAuth grants that you set up first.
Which means the agent never once touches the browser you actually do your work in. It carries its own credentials, its own service account, its own narrow keyhole view into Salesforce or Notion or whatever, granted through whatever integration the platform’s engineers happened to ship.
So before your shiny no-code agent does anything useful, you are wiring. Authorize this. Paste a key there. Map these fields to those fields. And then you cross your fingers that a connector exists at all for that one ancient internal tool your team can’t function without.
You already logged in this morning
This is the part that gets a little absurd when you say it out loud. You opened your laptop, signed into your CRM, your email, the analytics dashboard, the vendor portal with the genuinely terrible UI that you’ve stopped complaining about. Every one of those sessions is alive right now, sitting warm in your Chrome tabs, fully authenticated and waiting.
A cloud agent can’t see a single one of them.
It’s running on a server in some datacenter that has no idea you’re already standing inside the systems it’s trying to break into. So it knocks on a second door and asks you to authenticate all over again, this time on its behalf, which is a strange thing to ask a person who’s demonstrably already inside.
A browser-native agent just skips the whole performance. It operates inside the tab you’re already in, riding the session you established when you logged in. No connector to configure, no token to mint, no OAuth handshake to babysit. If you can see the page, it can act on the page. That’s the core idea behind Dassi, which lives as a Chrome extension in your side panel and reads precisely what you’re reading. We’ve made this argument before about why cloud browser agents can’t see your tabs, and honestly the StackAI deal is just the same blind spot wearing a no-code costume.
No-code was never the hard part
Drag-and-drop builders solve a real problem. Writing agent logic by hand is fiddly, and letting a sales ops person assemble a workflow without bugging an engineer is a genuinely good outcome.
But the friction in agents was never the code. It was the access.
The plumbing is the whole story
Strip away the marketing and the difference between these approaches comes down to one boring infrastructure question: where does the agent get permission to act, and how much of your day do you spend granting it?
The cloud model answers that with integrations. Each new tool means a new connector, a new scope to approve, a new place your data passes through on its way to the agent’s servers and back. Over a few months you’ve quietly assembled a sprawl of standing permissions you’ll never audit, half of which point at tools you stopped using in March, and every one of them is a credential someone could eventually misuse.
The browser model answers it differently, by inheriting the access you already have instead of demanding a parallel copy of it. Your logins stay yours. The agent borrows the session for the moment it’s acting and nothing leaves the machine you’re sitting at. With Dassi you can run that on your own ChatGPT subscription or bring your own key for Claude, Gemini, or DeepSeek, which matters because BYOK keeps your data between you and your model provider with no middleman quietly logging it. There’s no second set of credentials for an attacker to phish, because there’s no second set of credentials at all.
So I’ll be curious to see what Asana actually ships once StackAI is folded in. More connectors, probably. A slicker builder, definitely. A bigger catalog of the integrations they decided were worth building.
None of which changes the fact that the most powerful, already-authenticated, fully-context-aware environment you own is the browser tab open in front of you right now. The cloud agents keep trying to recreate it from scratch on a rented server, and damn if that isn’t the long way around.
If you want to try the short way, Dassi is on the Chrome Web Store. It’s free, and it can already see the tab you’re in.