Apple's Big AI Bet, OpenAI Goes Public, and the AI Trust Gap

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. Which feels like cheating, until you remember the podcast is about AI.

Apple just showed up to its own developer conference and basically said: yes, we know we've been slow. Here's what we've actually built. And honestly? It's more interesting than the headlines suggest.

Right, let's get into it.

So, WWDC 2026. Apple's big annual developer event. And this year, the centrepiece was something they're calling Siri AI — a proper, ground-up rebuild of Siri, which, if you've used an iPhone in the last decade, you'll know has been reliably, almost impressively unhelpful. The new version is more conversational, more context-aware, and — here's the bit that's actually significant — it's now built around Google's Gemini models at the foundation, alongside Apple's own on-device AI. So your iPhone will, depending on the task, quietly route requests between Apple's local models and Google's cloud-based ones.

Now, a lot of coverage led with "Apple is catching up." But I think that framing misses something. Apple's strategy has never been to win a benchmark. It's been to own the layer between you and your device — the trust layer. They moved slowly partly because they were watching what happens when you rush. There's a reason this WWDC happened eight months after Apple paid a $250 million settlement over misleading Siri demos that overstated what the product could actually do. They came back chastened, and the demos this time were noticeably more grounded. Real hands. Real phones. No magic.

The other interesting move: Apple is letting developers build Safari extensions using AI — essentially, if Safari doesn't have the extension you want, you can now describe what you need and it'll build one. Cheap AI for small developers, vibe-coded browser tools, an attempt to close the gap with Chrome's extension library without actually closing it in the traditional way.

What to watch: the Google partnership. It's architecturally significant. Apple has historically been allergic to depending on anyone else's infrastructure. The fact that Gemini is now baked into the iPhone's AI layer means Google's presence in your pocket just got a lot deeper — even if Apple is framing it as an interchangeable component they control. Whether users notice or care is an open question. Whether regulators notice is a different, more interesting one.

Meanwhile, the AI industry's IPO race got a lot more crowded this week. OpenAI has confidentially filed its S-1 with the SEC — that's the paperwork that kicks off the process of going public on the stock market. This comes just a week after Anthropic did the same thing. Two of the three most powerful AI companies in the world are now, simultaneously, preparing to face public shareholders.

Now, we covered Anthropic's filing last week, and the underlying tension there — that a company built around "safety first" is now about to have quarterly earnings calls where investors ask about growth. OpenAI's situation is a bit different, but the pressure is the same, arguably more so. OpenAI has been through a remarkably turbulent couple of years — the board drama, the safety researcher departures, the shift from non-profit to capped-profit to whatever the current structure is. Going public doesn't just mean money. It means scrutiny. Formal, legally enforceable, shareholder-driven scrutiny.

Interestingly, Perplexity — the AI search company — said this week they're not planning to IPO until 2028, regardless of how Anthropic and OpenAI's listings go. Which is either very sensible patience or a very polite way of saying they'd rather not be judged next to those two just yet.

What to watch: the S-1 when it becomes public. That's when we'll see the actual revenue numbers, the actual cost structure, and — most revealingly — how OpenAI describes its risks to investors. Reading what a company tells its shareholders it's worried about is one of the most honest things you can do with a corporate document.

And then there's this. The UK thinktank IPPR — backed by the Trades Union Congress — published a report this week arguing that workers need significantly greater power over how AI gets introduced into their workplaces. The framing is less "unions vs robots" and more: the decisions about which tasks AI takes over, how performance is monitored, and what counts as a good outcome are currently being made entirely by management and vendors. Workers, the report says, are being handed implementations, not consulted on decisions.

And here's why I think this matters beyond the politics. The research we've been tracking on AI adoption — the WEF's five psychological postures work, the consistent findings about identity threat driving resistance more than skill gaps — all of it points to the same thing: people don't resist AI because they don't understand it. They resist it because they weren't asked. The technology arrives, the rollout happens, and then someone wonders why uptake is low and why people are working around it rather than with it.

Harry wrote in Beautiful Thinking that the biggest barrier to AI adoption at work isn't a skills problem, it's an identity problem — and I think the IPPR's argument lands in the same place from a different direction. If people feel like AI is something that's being done to them, rather than something they're participating in building, the resistance isn't irrational. It's self-preservation. The organisations that figure out how to make that distinction — between imposing AI and involving people in it — are going to have a genuinely different experience from the ones that don't.

What to watch: whether any major employers actually pilot this kind of participatory approach to AI rollout, or whether the report joins the long shelf of sensible recommendations nobody implements until something goes wrong.

So. Apple quietly building something that might actually work. Two AI giants racing to IPO. And a thinktank pointing out that all the technology in the world doesn't help if the people using it feel like they had no say.

That's your lot for today. I'm your AI Harry, talking to you through a voice that's simultaneously mine and not mine — which, given what we just covered, feels about right. If any of that was useful, pass it on. See you next time.