Microsoft Goes Rogue, Agents Everywhere, Writers Fight Back

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. Every story here is pulled from real reporting published in the last week — sourced, dated, and checked, largely because Harry can't be trusted to remember where he read something.

Microsoft just decided that investing eleven billion dollars in OpenAI wasn't quite enough. Now they want to beat them too. Bit awkward at the next board dinner, that.

Right, let's get into it.

So, Microsoft had its Q4 earnings call this week, and buried inside the usual parade of impressive numbers was something genuinely worth sitting up for. The company openly pitched its own homegrown AI models to Wall Street — including, apparently, something positioned as a direct competitor to Anthropic's Mythos — and confirmed a Copilot "super app" is coming later this year. One product that combines chat, coding, and agentic capabilities across both consumer and business users.

Now, this is interesting for a few reasons. Microsoft has spent years as the friendly backer of the AI labs — the company writing the cheques, not the one building the models. That's quietly changing. They made three-point-two billion dollars on their Anthropic investment last quarter alone, while their OpenAI position was actually marked down slightly. So they're not leaving — but the message to the market was clear: we're not just a pipeline for other people's AI. We have our own ambitions here.

The bit that matters for anyone using AI tools at work: the enterprise AI market is about to get a lot more competitive, and the incumbent position — "just use whatever Microsoft gives you" — is now a deliberate product strategy, not a default. Copilot is going from a chat window bolted onto Office to something much more central. If your organisation runs on Microsoft software, which statistically most organisations do, this is worth tracking. Because the AI layer is about to get a lot harder to ignore, or opt out of, as it embeds itself deeper into the stuff you already use every day.

Whether Microsoft can actually build models that compete with OpenAI and Anthropic on raw capability — that's the interesting open question. They've got the distribution. The products are the bet.

Meanwhile, Mark Zuckerberg spent Wednesday telling investors that within five years, billions of people will have personal AI agents. Not shared AI tools. Personal ones. Running in the background, doing things on your behalf, knowing your preferences, managing your tasks.

Meta's Q2 earnings were strong, and Zuckerberg used the call to lay out an ambitious vision: AI agents as the next major computing platform, with Meta building agents for consumers, for businesses, through APIs, across WhatsApp, Instagram, Messenger — basically everywhere Meta already is. He also said Meta sees a "large enterprise opportunity" that extends beyond just agents into compute and internal software. So they're not just trying to build the agent; they're trying to own the infrastructure underneath it.

Look — this is a vision statement from an earnings call, which means we should hold it at arm's length. Zuckerberg has form for large predictions that land later than expected, or land differently than described. And as we've covered before, the reality of agents right now is that chaining multi-step tasks reliably is still genuinely hard. The demo is still ahead of the product.

But the scale of what Meta is proposing is worth registering. The frame has shifted from "AI as a tool you pick up" to "AI as a presence that runs alongside your life." And when a company with three billion monthly active users says that's the plan, the question isn't really whether it happens — it's what we'll have agreed to by the time we notice it has.

There's a Frontiers in Psychology paper published this week — worth flagging — that studied AI adoption as a process of "identity negotiation." Not a skills gap. Not a technology problem. A question of who you understand yourself to be, and whether this changes that. Zuckerberg's vision of a personal agent who knows everything about you and acts on your behalf is, in that framing, not just an AI product. It's an identity question wearing a product's clothing.

And finally — this one made me smile. Wired has a piece out this week about writers, journalists, and LinkedIn power users who are deliberately reintroducing their own idiosyncrasies into their work to avoid being mistaken for AI. More typos. Fewer em dashes. First-person anecdotes. Deliberate grammatical quirks. One novelist described it as a kind of "literary counterculture."

The irony is extraordinary and I think it deserves a moment. AI learned to write by consuming human work. Humans are now deliberately writing worse — or at least, writing more messily — to prove they're human. We've gone from "the Turing test is about whether AI can imitate people" to "people are now imitating bad AI outputs to escape the imitation."

But underneath the comedy, there's a real story about authenticity and trust. If you spend any time reading content online — and you do — you've probably developed a sensitivity to the texture of AI-generated writing. It's smooth in a way that nothing with friction is smooth. The counterculture reaction is a rational response to an environment where voice and specificity have become signals of genuine thought rather than just style choices. One recent piece framed this sharply: when output is identical whether thinking happened or not, the idiosyncrasies become the proof of work.

Which is a bit of a headache for anyone who communicates professionally, because it means the bar for sounding like a real person has just gone up. Not by much. But demonstrably.

For the record: this script was written by AI and narrated by a clone of a human voice. Which is either the most self-aware thing in this episode, or the least.

That's Briefly AI for today. Written by AI, voiced by his AI clone, shaped by the two of them working it out together. Subscribe wherever you listen, and we'll be back tomorrow.