Sam Altman wants your login credentials. Not just your email, but the keys to your bank account, your United Airlines frequent flyer portal, and that dusty Slack workspace you haven't checked since 2022.

OpenAI’s latest hire makes that ambition very clear. By bringing on Peter Steinberger, the mind behind OpenClaw, the company is signaling that it's bored of just being a smart-aleck text box. It wants to be an employee. Specifically, an employee that lives inside your browser and does the clicking for you.

Steinberger isn't your typical AI researcher. He doesn't spend his days staring at loss curves or debating the existential risks of a digital god. He’s a builder. A tinkerer. Before he started messing around with agents, he built PSPDFKit, a company dedicated to the soul-crushing but necessary task of making PDFs behave. He knows how to handle messy, legacy code. He knows how to make software actually work in the real world, rather than just in a pristine laboratory environment.

OpenClaw, his most recent claim to fame, is a tool designed to let AI "see" a web browser and interact with it like a human would. It clicks buttons. It scrolls. It fills out forms. It doesn't ask for permission; it just gets the job done. By swallowing Steinberger and his tech, OpenAI is effectively admitting that the "chat" part of ChatGPT was always just a front. A gateway drug. The real endgame is "Agentic AI"—a buzzword that translates to "letting a black box run your life."

We’ve seen this movie before. Every few months, a new startup claims they’ve built a "universal controller" for your computer. Then you try it, and it spends forty minutes trying to find the 'Buy Now' button on Amazon before hallucinating a 404 error and crashing your Chrome tabs. It's a hard problem. The web is a dumpster fire of bad UI and bot-detection scripts. Steinberger is being brought in to fix the plumbing.

But there’s a massive, expensive catch.

Letting an OpenAI-powered agent handle your chores isn't just a technical challenge; it’s a security nightmare. We’re talking about a system that needs to bypass two-factor authentication and hold onto your session cookies. If an agent can book a flight for you, it can also drain your savings account if it gets "prompt injected" by a malicious website. OpenAI hasn't really explained how they plan to keep these digital butlers from being robbed on their way to the grocery store.

Then there’s the compute cost. Running a massive language model just to navigate a crappy JavaScript menu is like using a SpaceX Falcon 9 to deliver a pizza. It’s overkill, and it’s pricey. Microsoft and OpenAI are already burning through billions in hardware. Now they want to add the overhead of "computer use," which requires constant screenshots and high-frequency processing. You can bet that $20-a-month Plus subscription won't stay at $20 for long if your agent is spending all day arguing with Comcast’s customer service bot on your behalf.

The friction here is palpable. Anthropic already has "Computer Use." Google is working on "Jarvis." OpenAI is late to the party, which is why they’re snatching up guys like Steinberger. They’re desperate to move past the "hallucination" phase where the AI just lies to you. They want it to be useful enough that you’ll overlook the fact that you’re handing the keys to your digital life to a company that can’t even decide if it’s a non-profit or a multi-billion dollar shark.

The industry is pivoting from "AI as a consultant" to "AI as a proxy." We’re being sold a future where we never have to look at a boring interface again. No more clicking through three pages of filters to find a hotel. No more manual data entry into Excel. It sounds great on a slide deck. It sounds like a dream for the overworked middle manager.

But there’s a certain irony in hiring a guy who spent years perfecting the PDF—the most rigid, unchangeable format in history—to help build the most fluid, unpredictable software ever conceived. We’re trading our autonomy for a few saved minutes and a lot of blind trust.

Is the goal to make us more productive, or just to make us more dependent on the subscription? Probably both. We’re moving toward a world where you don’t own your workflows anymore; you just rent a digital ghost to haunt your machine.

If Steinberger succeeds, you might never have to file an expense report again. But you’ll also have no idea how it actually got done, or who else was watching while the agent logged into your company portal.

Does anyone actually remember their own passwords anymore?