Microsoft Builds a Brain, Trump Blinks on AI Safety

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. The real Harry had the idea; the synthetic one is doing the talking.

Microsoft just announced its own AI model. And Trump just signed an executive order about AI safety — the one he shelved two weeks ago after a phone call from Elon Musk. Busy day. Let's get into it.

Right, let's start with Microsoft, because Build 2026 happened this week and there's quite a lot to unpack — I mean, quite a lot to go through. The headline announcement is something called MAI-Thinking-1. That's Microsoft's first in-house reasoning model — "reasoning" meaning it doesn't just predict the next word, it actually works through problems step by step before answering. Think of it like the difference between someone who blurts out the first thing that comes to mind versus someone who takes a breath and actually thinks. Microsoft has been powered by OpenAI's models for years, but the two companies recently renegotiated their relationship — loosening ties, as the polite framing goes — and MAI-Thinking-1 is the most visible sign that Microsoft is now serious about building its own AI stack rather than renting someone else's.

They also announced something called Microsoft Scout, which is — look, honestly, it's Microsoft's version of what Google has been doing with its OpenClaw-style assistants. Always-on, lives inside Teams and Outlook and OneDrive, can organise your calendar, draft emails, handle expense reporting. It shows up in your Teams workspace like a digital colleague who never sleeps and, unlike most colleagues, is entirely focused on doing the dull stuff you don't want to do. Which sounds like a gift, and probably is, for the dull stuff. The question nobody's quite answering yet is what happens to the people whose job *was* the dull stuff — but we'll come back to that in a moment.

Why does any of this matter? Because Microsoft is one of the most widely used enterprise software platforms on the planet. If you work in an office, you are almost certainly using Outlook or Teams or Word. Microsoft putting AI agents directly into those products — not as add-ons you have to seek out, but as default features embedded in the interface — means AI assistants are about to become ambient for millions of people who never chose to adopt them. The adoption curve just got flattened by distribution. Worth watching: whether Scout remains opt-in or becomes default, and how workers actually respond when the AI coworker shows up uninvited.

Meanwhile, on a rather different note — Trump signed an executive order on AI this week. Now, we covered a couple of weeks ago how his administration shelved a planned AI safety order after last-minute calls from Musk and Zuckerberg. That order, it seemed, was dead. Except it's not quite dead — it's been rewritten.

The new executive order creates what's being called a "voluntary framework" for AI companies to share their frontier models — their most powerful, most capable systems — with the federal government before those models are released publicly. The idea being that government can check for security vulnerabilities, risks to critical infrastructure, that sort of thing. Now, the word *voluntary* is doing a lot of heavy lifting there. Voluntary means companies can say no, and the current administration's track record suggests they probably won't be pushing very hard if they do.

So what's actually changed? Honestly, not much in practice. The US remains the only major economy without a mandatory pre-release review process for powerful AI. What has changed is the optics — the White House can now say it's doing *something* on AI safety without actually requiring anything of the companies it depends on for political and financial support. Whether that something becomes something more meaningful if, say, an incident forces the issue — that's the thing to watch.

The Wired framing this week put it well: the administration is at war with itself over AI regulation. There are officials who genuinely want guardrails; there are industry voices with direct lines to the Oval Office who don't. For now, the industry voices are winning. That may not last.

And now, the story I genuinely found most interesting this week — and it's a human one.

Uber capped its employees' AI spending. Mid-year. After burning through the entire annual budget in four months.

Here's what happened: Uber, like a lot of companies right now, told its staff earlier this year to use AI as much as possible. Encouraged it, enthusiastically. And so they did — extensively, expensively — to the point where Uber had to step in and put a lid on it. Individual spending limits, approval processes, the works.

Now, on the surface this is a finance story. But I think it's actually something more interesting than that. It's evidence that when you genuinely remove the friction from AI adoption — when you say "use it for everything, it's free, go for it" — people actually do. Heavily. The question of whether AI will change how we work has, for a lot of companies, been a theoretical one. Uber just ran the experiment by accident and got their answer: yes, quite dramatically, and faster than the budget forecast anticipated.

There's a lesson here for any business thinking about AI rollout. Unlimited access is not a sustainable model, as GitHub Copilot also worked out recently when it moved from flat subscription to per-token billing. The "AI for everything, figure out costs later" era is ending. What's replacing it is metered, managed, and a lot more like a utility bill than a software subscription. If you're in procurement or finance, this is the thing to watch — because the AI spend conversation is landing on your desk whether you're ready for it or not.

That's your lot for today. Microsoft built its own brain, Trump signed a voluntary safety order that doesn't require very much, and Uber discovered that if you tell people AI is free, they will use a lot of AI. Three stories, five minutes, no existential dread. If any of that was useful, tell someone. See you next time.