Transcription
Last week, Sam Olman made the most expensive talent hire in OpenAI's history. He brought on Peter Steinberger, the creator of OpenClaw, the fastest growing open-source project ever built. Open Claw, Open Clay, Open Claw, Open Claw, >> Open Claw. >> Every headline ran the same story. Talent war, Alman beats Zuckerberg, standard tech drama, but nobody connected it to what's been happening inside Goldman Sachs for the past 6 months or to the enterprise platform Alman launched. exactly one week earlier.
Once you see the full picture, you'll understand why this single hire could accelerate job losses faster than any AI development in history. So, what exactly did Altman get his hands on? OpenClaw isn't a chatbot. It's not chat GBT in a different rapper. It's an autonomous AI agent that takes control of your entire computer. It reads your emails. It writes and runs code. It browses the web. It books meetings. It manages your files. It can even negotiate on your behalf through any platform, WhatsApp, Slack, email, whatever.
And the stories coming out of the community are absolutely insane. One user set his OpenClaw agent loose on a car dealership. The agent went back and forth over email, haggled the price, and negotiated a $4,200 discount. The hype has been so big that it literally caused a Mac Mini shortage across the United States as people kept buying hardware to run it 24/7. Peter Steinberger built all of this alone from his apartment in Austria. Meta and OpenAI both made offers. Steinberger described them as billion-dollar corporate buyouts on Lex Friedman's podcast. Zuckerberg personally reached out. He also spoke with Satya Nadella from Microsoft, but Alman won. And Steinberger's own prediction, now effectively OpenAI's road map, says everything you need to hear. Openclaw style agents will kill 80% of apps. And so when computers can just do all these things, are apps just going to go away? >> Uh, I think 80% of them are going away. >> Why would Alman take such a big bet on one person? What does he know that we don't?
To answer that, we need to talk about something almost nobody has reported on. For the past 6 months, with almost zero press coverage, Anthropic engineers have been embedded inside Goldman Sachs. Not selling them software, not running a chatbot pilot, actually embedded inside their offices, building autonomous AI agents from the inside. Think about that for a second. One of the leading AI companies in the world sent their own engineers to physically sit inside one of the most powerful financial institutions on Earth and build agents that do the work of Goldman's employees. And these agents don't answer questions, they complete work.
Here's what that actually looks like. A trade happens. Millions of transactions need to be reconciled across systems. So the AI agent pulls the data, cross-references it across the system, flags discrepancies, and then files the settlement. No human needs to touch it. So if a new client walks in, the agent passes their passport, cross-references global compliance databases, runs KYC and anti-money laundering checks, triggers the right workflows, and generates the onboarding file. A process that used to take a team of analysts days, now done in minutes. Goldman CIO doesn't call these tools. He calls them digital co-workers. Goldman has 47,400 employees. Thousands of them do exactly these jobs. And their internal strategy memo, signed by the CEO, calls for limited reduction in roles. Their CFO put it even more bluntly. This is a fundamental rethinking of how we expect our people to operate at Goldman Sachs.
Here's the thing about Anthropic. Their sweet spot has always been on the technical end. Developers, researchers, big institutions that want precision and control. Their biggest deals happen behind closed doors with companies like Goldman. Altman doesn't operate like that. When it comes to products, OpenAI ships to everyone first and figures out the business model later. ChatGPT wasn't pitched to boardrooms. It was dropped on the internet and hit 100 million users in 2 months. That's the opposite of how Anthropic operates, and it tells you exactly what Alman plans to do with autonomous agents. So when Altman sees Goldman spending 6 months and millions of dollars building custom AI agents with Anthropic, he doesn't want to compete for that contract. He wants to make that contract unnecessary. That's why he hired Steinberger. And that's why one week before the hire, he launched OpenAI's enterprise agent platform called Frontier.
The playbook is not subtle. Take what Goldman built behind closed doors, productize it, ship it to every business on the planet, and the infrastructure is already forming. Nine startups in Y Combinator's latest batch are building enterprise products on top of OpenCore right now. One company called Tensil is deploying what they call AI employees, agents that plug directly into your company's Slack, HubSpot, GitHub, and Gmail on day one. Another called Bits offers a security-hardened version with 3-minute setup. Think about what this means. The same agent that reconciles Goldman's trades can reconcile a small firm's books. The same agent that runs Goldman's compliance can do a law firm's client intake. The same agent that generates Goldman's pitchbooks can write a marketing agency's proposals. Not as well, at least not yet, but at 1/100th of the cost. Not as well is more than good enough for most CEOs to pull the trigger. Goldman built the proof of concept. Altman is allowing the entire world to have access.
So, here's where this stops becoming a tech story and starts becoming everyone's problem. Goldman Sachs, the same Goldman that's building these agents behind closed doors, published research estimating that 300 million jobs worldwide are exposed to AI automation. 300 million. The IMF went even further, citing 40% of all global employment. In the US alone, 2/3 of occupations have meaningful exposure. But here's the part that nobody is processing yet. Every previous technological revolution – the printing press, the steam engine, electricity, the internet – they all just placed workers at the bottom first. Factory workers, manual laborers, the people doing physical repetitive tasks. AI is doing the opposite. Goldin's own research found that AI can automate 46% of administrative and office tasks, but only 1% of maintenance and manual labor. 46 to 1. Accountants before janitors, lawyers before electricians, analysts before plumbers. The higher your salary, ironically, the more exposed you are. For the first time in the history of automation, it's the most educated, highest-paid workers who are at the most risk. And that is a fundamentally different kind of disruption than anything we've ever seen before. Because the people being displaced this time are the people who write the laws, manage the money, and run the institutions.
And right now, this is the number that should keep you up at night. 94% of companies haven't even started deploying AI agents yet. We are at the very beginning. Goldman proved that autonomous AI agents work in the most complex, regulated, high-stakes environment on the planet. Sam Alman just hired the person who can make that capability simple enough for every single business on Earth. And the startups are already building the plug-and-play versions. The question was never whether this would happen. It was always when, and the answer just walked through OpenAI's front door.