I read a Klarna earnings note this week where the CEO quietly walked back the company’s famous “we replaced 700 customer service agents with AI” line from 2024. Turns out they’re hiring humans again. And the AI did the job, technically, but customers hated it and revenue suffered.

This keeps happening. Companies make a loud claim about AI replacing workers, then walk it back six months later, and nobody updates their forecast. So when I see Goldman Sachs predicting 300 million jobs displaced or Anthropic’s CEO suggesting 50% of entry-level white-collar jobs vanish by 2027, my read is that they’re directionally right and specifically wrong.

The future of work is changing. But not in the way the headlines say.

What’s actually shrinking

The work surface. The thing knowledge workers do their job through. For most people that used to be a desk, a phone, and maybe an email client. But then the desk got smaller, the phone moved to Slack, and the email client became one tab in a browser holding sixteen other tabs.

Now it’s just the browser. The thing you’re reading this in. So that’s where the work happens — Salesforce, Notion, Gmail, Linear, Figma, every dashboard, every form, every approval workflow. The work surface didn’t disappear when companies got remote-friendly. It collapsed into Chrome.

And if that’s where the work happens, then AI that doesn’t live in the browser is AI that doesn’t live where the work is. Which explains a lot of the productivity numbers I keep seeing in 2026.

The tab count problem

Atlassian’s December 2025 report had the average knowledge worker at 14 tabs open during the survey. And it’s probably worse now. That’s not productivity. It’s evidence of a job nobody designed.

A prediction I’ll commit to

By 2028 the job posting for “Operations Manager” or “Customer Success Lead” or whatever middle-of-the-org role you want will list “browser agent fluency” as a requirement. Not “AI fluency” — that’s already on every posting and means nothing. But browser agent fluency specifically. Because the work happens in the browser, and the difference between a person who can compose a five-step browser agent and one who clicks every step manually is the difference between closing 40 tickets a day and closing 8.

So this is the same shift that happened when Excel went from spreadsheet to required job skill. Nobody hires accountants who can’t use Excel anymore, and nobody asks them to defend that in an interview. And in ten years nobody will hire ops people who can’t compose an agent to do their work. The interview question won’t be “do you use AI.” It’ll be “show me a workflow you automated last week.” The companies that move first will quietly run leaner teams while the laggards keep posting 30-person teams to do what 8 augmented people can do. But that gap won’t show up in McKinsey reports. It’ll show up in margins.

And the thing nobody wants to say is that this also changes who gets hired. Browser agent fluency is not correlated with elite credentials. It’s correlated with curiosity and a willingness to mess around with tools. Some of the best agent composers I’ve talked to never finished a CS degree. Or they did, ten years ago, and forgot most of it. Hell, some of them are running ops teams now. Some of the worst are senior PMs at big companies who still think “AI” means “I prompted ChatGPT once.”

Where this leaves dassi

We built dassi on the bet that the browser is where work actually happens, and any AI tool that requires you to leave it is going to lose to one that doesn’t. So the future of work doesn’t run through a cloud browser sandbox. It runs through your already-logged-in tabs, with the LLM of your choice, on the schedule you set. We wrote about why cloud browser agents miss the point a while back, and the gap has only widened since.

So my prediction, for what it’s worth: in three years the question won’t be “remote or office.” It’ll be “does your team use agents or not.” Because the office vs. remote debate was about location, and the next debate is about what runs in your tabs while you’re not looking. And whichever team wins that one will be the one that stops measuring presence and starts measuring the things their agents queued up overnight.

Klarna will hire some of them back too, probably.