Dara Khosrowshahi just casually mentioned on Steven Bartlett’s podcast that Uber engineers built an AI version of him. A chatbot clone of the CEO, trained so teams could rehearse presentations before walking into the real meeting.

The Hacker News thread blew up. And I get why, it’s a wild image. Pitching your quarterly plan to a robot version of your boss, tweaking your slides based on what the algorithm thinks Dara would ask.

But something about the way people reacted to this bothers me.

The direction matters

Most of the commentary focused on whether AI should be replacing managers. Top-down automation, corporate efficiency, all that. Interesting debate. But it completely misses what’s actually useful about AI for the other 99% of us who do not report to Dara Khosrowshahi.

Because the real question isn’t “should companies build AI versions of their executives.” That’s an Uber problem. A Fortune 500 problem. An organizational theater problem, frankly.

What matters to most knowledge workers is way more boring: who is the AI working for?

Two very different directions

Uber’s version is top-down. The company decides to build it. The company controls what it knows. The company benefits because meetings presumably run smoother and engineers come more prepared.

But there’s another direction entirely, and it’s the one I find way more interesting because it actually puts individual workers in control of their own day-to-day productivity instead of optimizing them for someone else’s strategic agenda. Bottom-up AI automation is when you pick the tool, you decide what it does, and the results go straight to making your own workday less miserable.

Top-down AI automation optimizes workers for the company. Bottom-up AI automation optimizes the company’s tools for the worker. And right now, bottom-up is winning in practice even if it gets less press than Uber’s CEO chatbot.

2pm on a Tuesday

But I keep coming back to this. The big AI stories are always about strategy, executives, organizational transformation. Meanwhile the actual daily pain for most people is filling out the same vendor form for the fourth time this month, copying data between tabs, writing yet another follow-up email that’s 80% identical to the last one. And that’s the stuff that eats your afternoon. Not a lack of AI-powered management wisdom.

The HN discussion had a running theme: people sharing stories about their companies spending months building fancy internal AI tools for “strategic alignment” while everyone on the actual team was still manually entering the same damn data into three different systems every single day. Classic top-down thinking. You automate the executive’s pain point and completely ignore the grunt work that burns hours for the people doing it.

Browser agents flip this around. dassi sits in your Chrome side panel, sees what you see on screen, and handles the repetitive crap without you needing to copy-paste context into some separate chat window. You stay in your browser where your actual work already lives. The AI works for you, not for your boss’s meeting prep. And because you bring your own API key, the company does not even need to be involved.

Uber’s 90% stat is the interesting part

Khosrowshahi mentioned that about 90% of Uber’s engineers now use AI tools, with roughly 30% being “power users.” And that tracks with what I’m seeing more broadly. But adoption is not the problem anymore.

Most of those engineers aren’t using the Dara chatbot. They’re using coding assistants, debugging tools, things that make their individual work faster. Bottom-up tools. The CEO clone is a fun story but it’s probably the least impactful AI deployment at the entire company.

So when Khosrowshahi says AI is “changing their productivity in a way that I’ve never seen before,” I believe him. I just think the productivity gains are coming from developers using AI to automate their own grunt work, not from rehearsing with a robot boss.

The gap nobody talks about

Developers got bottom-up AI tools early. Copilot, Cursor, Claude Code. And they’re years ahead of everyone else in figuring out how to delegate the boring parts of their jobs to machines.

But everyone else is still mostly stuck with ChatGPT in a separate tab, doing the copy-paste dance between their work and their AI. Or they’re waiting for their company to roll out some top-down initiative that may or may not actually help with their day-to-day.

But that gap is closing. Browser agents like dassi are doing for knowledge workers what coding assistants did for developers, which is basically saying “here, let me handle the parts of your job that don’t require you to actually think.” And the people who figure this out first are going to have a hell of an advantage over the ones still manually filling out forms and writing emails from scratch.

Dara AI is a symptom

Entertaining story. But the future that actually affects your Tuesday afternoon is quieter, and involves a lot less rehearsing with robot executives.

If you want to try AI that works for you instead of the other way around, dassi’s free and sits right in your browser. Beats a chatbot clone of your CEO, probably.