Microsoft shipped Scout this week, and the framing was exactly what you’d expect: an OpenClaw-inspired personal assistant that books your travel, triages your inbox, and “works alongside you.” I read the launch post and then scrolled through the Hacker News thread, which by Tuesday afternoon was sitting near the top of the front page next to three separate posts about autoscaling cloud browser agents and yet another agent leaderboard. The whole feed had a theme. And nobody seemed to notice they were all describing the same thing wearing different logos.

Scout is a cloud agent. So is most of what trended yesterday.

The model isn’t the interesting part anymore

Everyone wants to talk about which model is under the hood. Scout reportedly runs a tuned frontier model, the leaderboards rank everyone on task accuracy, and the comments argue about reasoning depth like it’s 2024. I get the instinct, because for a while the model genuinely was the bottleneck.

But that stopped being true. The frontier models are all good enough to read a page, plan a few steps, and click the right button most of the time. Once that’s commoditized, the question that actually decides whether the thing is useful is boring and unglamorous: where does the agent execute? On a server in some datacenter, or in the Chrome window you have open right now with forty tabs and a logged-in session for every tool your job touches.

What “cloud” quietly costs you

Here’s a concrete failure I keep watching people hit. You ask Scout to pull the latest invoices from your billing dashboard and summarize them. Scout spins up a clean browser in the cloud. That browser has never seen your dashboard. It has no cookie, no session token, no saved 2FA approval. So it lands on a login wall, and now you’re either re-authenticating inside Microsoft’s sandbox, or handing an autonomous agent your password and your second factor so it can do it for you, which is the kind of thing that sounds fine until you say it out loud.

Your actual browser already solved this. You logged in weeks ago. The session is sitting there, warm, authenticated, one tab away. A cloud agent cannot reach across the internet and borrow it, because that session lives on your machine and was never designed to be portable, and honestly that’s a good thing for security even though it’s the exact reason cloud agents keep faceplanting on anything behind a login. I wrote more about why cloud browser agents can’t see your tabs if you want the longer version.

Same plumbing problem, third time this quarter

Microsoft has done this before. They cloned the OpenClaw pattern, shipped a cloud agent, and called it new. We covered the last round when Microsoft built another cloud agent that still couldn’t see your browser. Scout is the same architecture with better marketing and a friendlier name.

Where browser-native changes the math

A browser-native agent runs as an extension inside the browser you’re already using. That’s the entire difference, and it’s bigger than it sounds.

Dassi lives in the Chrome side panel. When you ask it to grab those invoices, it’s already standing inside your authenticated session, because it’s literally part of the browser that holds the session. No re-login. No credential handoff. No sandbox that’s never met your tools. It sees the rendered page the same way you do, acts on it where you are, and the page content stays on your machine except for whatever you send to the model you picked. You can run it on your ChatGPT subscription or bring your own key for Claude, Gemini, or DeepSeek, so the “which model” debate becomes a setting you flip, not a vendor you’re stuck with.

The cloud version is genuinely better at one thing: it’ll keep running after you close your laptop. For long unattended jobs that don’t need your logins, that matters. But the moment a task touches something you’re signed into, like your email, your CRM, your bank, or your internal admin panel, the cloud agent is back at the login wall and the local one walked right in.

So when the next leaderboard drops and ranks Scout three slots above something else on raw accuracy, I’d ask a different question before caring about the score: can it see the page I’m actually looking at? Scout can’t. If you want one that can, Dassi is in the Chrome Web Store, free, and it’ll be logged into everything before you finish reading this.