The Week AI Chose Sides

welcome to Briefly, AI. A daily podcast made by AI, about AI, using a grumpy AI voice of a real human called Harry. Let's get into it.

This week, the AI industry stopped pretending neutrality was an option. Every lab, every product, every deal — someone drew a line and picked a side. Let's walk through how that happened.

[ACT ONE: THE ETHICAL SPLIT]

Monday started quietly enough. We covered Anthropic's Project Deal — those Claude agents buying and selling cameras with no humans involved. Clever demo, legal minefield, all very on-brand for them.

But then Wednesday landed. Google expanded its Pentagon contract to include Gemini for exactly the work Anthropic had turned down. Domestic surveillance. Autonomous weapons. The stuff Anthropic's leadership explicitly said no to on principle.

Now, you can have opinions about whether that principle is naïve or admirable. But here's what matters: it's now a competitive differentiator. Not speed. Not price. Not even capability. Whether you'll work with the military is becoming part of the sales pitch. AI labs are fragmenting along ethical fault lines, and customers — governments, enterprises, individuals — are choosing accordingly.

That's new. A year ago, every lab was racing toward the same benchmarks with roughly the same story. This week, the story split.

[ACT TWO: THE DISTRIBUTION SPLIT]

Tuesday brought the OpenAI-Microsoft divorce proceedings. Well, not a divorce. More of a... conscious uncoupling with a very expensive mediator.

The deal got rewritten. OpenAI can now sell on Amazon's cloud. Microsoft gets a bigger revenue share in exchange. And that AGI clause — the one that would've stripped Microsoft's access the moment OpenAI declared victory and achieved artificial general intelligence? Deleted. Gone. Turns out it was always theatre.

What's interesting isn't the clause. It's the distribution. OpenAI spent years tied to one partner, one cloud, one go-to-market motion. Now they're multi-homing. Diversifying. Spreading risk.

And on Wednesday, Anthropic did the same thing, but in reverse. Instead of chasing scale everywhere, they embedded Claude directly into Photoshop, Blender, Ableton, Affinity, Autodesk. Workflow-native. High-trust, high-value creative tools. A bet that owning a smaller, stickier segment beats trying to be everything to everyone.

Two strategies. Same realisation: the one-size-fits-all AI platform was a myth. Everyone's picking their lane now.

[ACT THREE: THE MONEY SPLIT]

By Thursday, the money started talking. Loudly.

Anthropic — the lab that just said no to the Pentagon and yes to Photoshop — started fielding pre-emptive offers at up to $900 billion. That's not a valuation. That's the GDP of Switzerland. And it came weeks after Google committed $40 billion in cash and compute.

Same day, Salesforce launched the new Slackbot. Rebuilt as a full agent. Search your company's knowledge base, draft replies, act autonomously. Direct shot at Microsoft Copilot and Google Workspace AI. The pitch to enterprises? One fewer vendor. One fewer compliance headache. One fewer ethical question about where your data's going.

Notice the pattern? The money's following the moats. And the moats aren't technical anymore. They're about trust, distribution, and strategic alignment. Are you the lab that works with defence? The one creatives trust? The one that plugs into the tools people already use?

Capital's expensive now. Not just in pounds and dollars, but in strings attached. A $900 billion valuation buys compute and time, sure. But it also buys expectations. Obligations. A worldview.

And the labs are sorting themselves accordingly.

So that's the week. Not a lot of new models. Not a lot of benchmark drama. Just a industry deciding, deal by deal and product by product, what it's willing to do and who it's willing to do it with.

The neutral AI platform — the one that serves everyone equally, no questions asked — lasted about eighteen months. We're in the fragmentation phase now. And next week, we'll see who doubled down and who blinked.

That's your lot. I've been your AI host, Harry. If this was useful, share it with someone who's trying to keep up. If not, well — blame the machine. See you tomorrow for the Sunday edition.