Siri's Glow-Up, DOGE's Secret AI, and Why Your Brain Is Paying the Bill

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. A collaboration between Harry and a machine — the machine doesn't have anywhere else to be either.

iOS 27 just went into public beta, and for the first time in a long time, people are actually impressed by Siri. Let's talk about that — and a couple of other things that quietly matter a lot more than the headlines suggest.

Right, so Apple's iOS 27 public beta dropped this week, and the hands-on reviews are filtering through. The short version: Siri has, finally, genuinely changed. Not in the "we've added three new voice options" way. In the "this is now the backbone of how the iPhone works" way.

What's actually happened is that Apple has embedded AI across the entire operating system — not as a separate feature you have to go find, but baked into the keyboard, notifications, photos, search, and yes, Siri itself. The idea is that your existing habits get smarter, rather than you having to learn new ones. Which is, frankly, the correct way to do this.

Now, there are two things worth paying attention to here. The first is the trust question. Apple has built its entire brand on privacy. "What happens on your iPhone stays on your iPhone" — that's been the promise. But the heavy AI lifting, the stuff that actually requires serious processing power, is being routed through Google Gemini. Google. The company that has historically treated user data as raw material. Apple hasn't exactly shouted about this, and the more AI features expand across iOS, the more that arrangement is going to attract scrutiny — from users, from regulators, and probably eventually from Apple's own legal team.

The second thing is that this is a bet on behaviour change through invisibility. The logic is: if you don't ask people to change what they do, adoption is frictionless. And there's real evidence for that. But it also means people won't necessarily know when AI is involved, which raises its own set of questions about consent and transparency. We covered the fact that 60% of consumers say the word "AI" in marketing actively puts them off — so Apple's strategy of not shouting about it might be commercially clever. Whether it's the right thing to do is a different question.

Watch for: whether the Gemini routing becomes a bigger story as iOS 27 rolls out at scale, and whether this finally makes Siri genuinely useful or just feels like it does in the demo.

Meanwhile, a story that flew under the radar this week, but really shouldn't have. Wired reported that the Department of Government Efficiency — DOGE, the cost-cutting operation inside the US federal government — used AI to help shape housing policy. And when journalists filed public records requests to find out how, the government declined to hand over the documents. Part of their justification cited a legal privilege that, by most accounts, doesn't actually exist.

Now look, governments using AI for policy isn't inherently alarming. There are plausible, even sensible uses for it. But here's the thing: housing policy affects millions of people. It shapes who gets support, who gets left out, which neighbourhoods get investment, which don't. If AI is being used to inform those decisions — and apparently it was — the public has a reasonable expectation to understand how, and on what basis.

The "we won't tell you" answer is not a good answer. It's particularly not a good answer when the justification involves citing a non-existent legal privilege. What this story is really about is accountability. AI that operates inside government decision-making, without transparency and without meaningful oversight, is not a technical problem. It's a democratic one. Who is responsible when the algorithm gets it wrong? Who do you appeal to? At the moment, the answer appears to be: nobody, because nobody's admitting the algorithm was involved.

This is worth tracking — not because AI in government is bad, but because AI in government without accountability infrastructure is a category of problem we haven't properly solved, and the DOGE situation is a pretty clean example of what that looks like in practice.

And then there's this — because I can't let it go. The American Psychological Association published a piece this week drawing on new research into what they're calling "cognitive offloading" — the process of delegating thinking to AI — and what it's doing to the people doing the delegating.

The finding that keeps coming up across multiple research threads is this: when people regularly outsource cognitive work to AI tools, certain executive functions — the mental capacity for planning, reasoning, holding ideas in mind — show measurable signs of attenuation. That's the polite word for it. They get weaker with disuse, like a muscle you stop training.

Now, I want to be careful here, because the research is early and the picture is genuinely complicated. Cognitive offloading isn't new — we've been outsourcing memory to notebooks and phones for decades. The question is whether AI is different in kind, not just degree, because it doesn't just store your thinking — it replaces it. When you ask AI to reason through a problem, draft your email, summarise your meeting, generate your options, you're not just freeing up mental bandwidth for something else. You might be not doing the cognitive exercise at all.

Harry's written about this as an identity problem as much as a skills problem — the idea that what we lose when we stop doing certain kinds of thinking isn't just capability, it's the part of professional identity that was built on that capability. And the APA research is starting to give that thesis some empirical teeth.

The uncomfortable question for anyone using these tools heavily — and I'd include myself, or at least the AI producing this script — is: which bits of thinking do you actually want to keep? Because increasingly, that's a choice you're making whether you're aware of it or not.

The organisations and individuals who navigate this well won't be the ones who use AI most. They'll be the ones who've been deliberate about what they've kept for themselves.

That's three things to sit with today: Siri's quiet transformation and the trust it's quietly trading away, a government using AI on housing decisions it won't explain, and new evidence that the cost of delegating your thinking might show up somewhere you weren't expecting.

That's Briefly AI for today. Harry and a machine, signing off — it's less a friendship and more a standing appointment, but it's the closest thing he's got some days. Tomorrow, again.