AI Agents Are Booking Appointments and the Pentagon Has ChatGPT Now

This was one of those weeks where three separate AI stories hit the same nerve from different angles — and if you run any part of your business with AI tools, all three matter.

ChatGPT can now sign into websites for you

OpenAI rolled out an update to ChatGPT Work — its agent mode designed to handle tasks autonomously. The big change: it can now sign into websites to complete actions like booking a DMV appointment, canceling a reservation, or filling out a job application. According to The Verge’s coverage, OpenAI says it does this without the company ever seeing your login credentials.

This is a significant jump from “type a prompt, get text back.” We’re now in the territory where AI handles multi-step web interactions that require authentication — the kind of tasks that previously needed you to be sitting in front of a screen. If you’ve been building automations with Make or Zapier to handle form submissions and bookings, this changes the calculus. The AI doesn’t need your automation anymore. It just… does it.

I covered how AI agents are becoming employees and what ChatGPT Work’s autonomy means for solo builders. The pattern is accelerating: the tools are getting from “suggests what to do” to “does it without asking.”

What to watch: If ChatGPT can sign into websites, the next question is what happens when it signs into the wrong ones, or fills out forms incorrectly. The Cursor situation where OpenAI cut off model access shows that platform control is real — and when the AI has your credentials, that control extends to your accounts.

The Pentagon now has custom ChatGPT and Grok

The Department of Defense announced it has added custom versions of ChatGPT and Grok to its GenAI.mil platform for unclassified military use. According to the Pentagon’s release, the portal launched last year with Google’s Gemini and has already onboarded more than 1.7 million of the department’s 3 million personnel.

This is worth paying attention to not because of the military angle, but because of what it signals about AI adoption speed. When an institution as bureaucratic as the Pentagon gets 1.7 million people onto an AI platform in under a year, the “we’re still evaluating AI” excuse in the corporate world doesn’t hold up anymore.

If you’re a solo builder wondering whether AI tools are “ready” for business-critical work — the answer is that the US military thinks they are. The question isn’t whether to adopt. It’s whether you’re adopting fast enough.

Even Microsoft’s Office chief is tired of AI slop

Ryan Roslansky, who leads Microsoft’s Office organization (and oversees LinkedIn), posted a warning on LinkedIn last week about AI-generated “slop” infiltrating daily work. He’d received a fully AI-generated document from a colleague and wasn’t happy about it. LinkedIn, which recently added an AI slop reporting button, has seen over a million people use it.

This is the most important story of the three for anyone using AI tools professionally. The problem isn’t that AI can’t write — it’s that people are using it to write things they should be thinking through themselves. A fully AI-generated document isn’t a productivity boost. It’s a signal that someone isn’t doing their job.

I covered the AI groupthink problem and how solo builders can avoid it. The tool is powerful. The failure mode is laziness. If your AI output looks like everyone else’s AI output, you’ve just turned your business into a commodity.

The practical takeaway: Use AI for research, first drafts, data analysis, and automation. Don’t use it to replace thinking. The Microsoft exec who builds AI tools for a living is telling you the same thing your gut already knows — if the document reads like AI wrote it, it shouldn’t have been written by AI.

What else happened

Google wants to license Disney characters for AI tools. Google is pitching studios on using AI for production and visual effects, but unions and creators are pushing back on copyright and job displacement concerns. If this goes through, it changes how content gets made at scale.

Google AI Mode will let you book hotels. Rolling out in the US in the coming weeks, with support for Expedia, Booking.com, Hilton, Marriott, and others. You’ll also be able to track flights in points by linking airline rewards. The AI tool ecosystem keeps expanding into real-world transactions.

Amazon is adding 2 million more Nvidia GPUs to AWS. On top of the 1 million already planned, because demand “exceeded expectations.” Nvidia posted record revenue again. The AI infrastructure buildout is not slowing down — which means the tools you use will keep getting faster and cheaper.

OpenAI’s advertising business hit $1B annualized run rate. Roughly 200 days after launching. OpenAI is also testing outcome-based pricing that charges only when AI agents succeed. If that model spreads, it changes how you evaluate AI tool costs.

The bottom line

Three stories, one pattern: AI is moving from “tool I use” to “agent that acts on my behalf.” ChatGPT signs into websites. The Pentagon deploys AI to millions of employees. AI-generated documents are flooding workplaces fast enough that the companies building the tools are telling people to stop.

If you’re building with AI, the question isn’t whether to adopt — it’s how to adopt without becoming the person who sends AI slop to their colleagues. Start with the AI Tool Advisor if you’re picking tools, or check out how to build your first automation if you’re just getting started.