Erik Brynjolfsson told Bloomberg the AI productivity takeoff is “finally visible in the data.” And enterprise AI adoption crossed 70 percent in Deloitte’s latest survey. But a McKinsey survey from late 2025 found the median knowledge worker saves less than 30 minutes per week with AI tools. Thirty minutes. Models ace graduate-level reasoning exams, write production code, summarize 50-page legal briefs in seconds, and the average person’s weekly time savings wouldn’t cover a lunch break. Robert Solow’s old quip about seeing computers everywhere except the productivity statistics has curdled into something uglier, because now the tools are obviously, embarrassingly capable and we’re still just ferrying text between browser tabs like carrier pigeons with clipboards.

Where the time actually goes

Gloria Mark’s research group at UC Irvine measured the average knowledge worker switching between applications 1,200 times per day, with each context switch on a complex task carrying a cognitive recovery cost of roughly 23 minutes. So AI chatbots didn’t help with any of that. They bolted another tab onto the pile.

A recruiter on Reddit last month laid out her “AI-powered” workflow: open LinkedIn, find a candidate, copy their profile summary, paste into ChatGPT, ask for a personalized outreach draft, copy the output, switch back to LinkedIn, paste, tweak, send. Nine steps and four tab switches, repeated forty times a day. And sure, the drafts were better than anything she’d write cold. But she was still spending most of her energy as a human copy-paste buffer, shuttling text between apps that refuse to share context, rebuilding her train of thought after every switch, losing minutes in the cracks between windows. Because the bottleneck was never the writing. It was the clicking. And she’s not unusual. Sales reps do the same dance between their CRM and a chat window. So do support agents toggling between Zendesk and an AI summarizer. So does basically every knowledge worker whose tools live in separate tabs that share nothing with each other.

Browser agents skip all of that

A browser agent lives inside your browser, sees the page you’re on, acts on it directly. So there’s no copying, no pasting, no alt-tabbing to a chat window. It reads the Gmail thread itself before drafting a reply. And it pulls from context you already gave it when filling a procurement form. Or it opens competitor pages, reads them, compiles notes without you orchestrating each step through a separate interface.

The 7 tasks you’re still doing manually in your browser are 90-second chores that happen 30 times a day. And nobody was ever going to write a Python script for something that minor. So these tasks sit in a dead zone that traditional automation ignores completely.

Two things changed

  1. Multimodal models learned to see web pages. Old browser automation leaned on DOM parsing and CSS selectors that shattered every time a site pushed new markup. So modern agents look at a page the way you do, visually, identifying buttons and forms by appearance, which dragged reliability from “demo that breaks on the second try” to something you can leave running unattended.
  2. API costs collapsed. Claude 3.5 Sonnet, GPT-4o, Gemini 2.0 Flash. In 2024 having an AI read and reason about a full webpage cost $0.50 or more per interaction. Now it’s pennies. And that price drop is what made continuous browser-level AI assistance a real product instead of a research curiosity.

Most people don’t know this category exists

One user in our community cut daily LinkedIn outreach prep from 3 hours to 40 minutes after switching to AI-assisted LinkedIn workflows without copy-paste. Two hours and twenty minutes back, every day. But most people still equate “AI for work” with typing prompts into ChatGPT. They’ve settled into the copy-paste loop because it feels productive enough — the way hand-washing dishes feels productive enough until you finally buy a dishwasher and realize you’d been wasting an hour every night for years. So the productivity numbers will keep climbing unevenly. Because gains go to whoever stops treating AI as a search engine trapped in its own tab.

Put AI where the work is

dassi sits in your browser’s side panel and handles drafting, research, form-filling, and repetitive tasks right on the page you’re already looking at. Bring your own API key. Your data stays local.