Sandwiches, Sovereignty, and a Very Honest CEO

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.

Right. Jersey Mike's filed to go public this week. And somewhere in their IPO documents — a sandwich company, I want to be clear — someone wrote the word "artificial intelligence." Not once. Multiple times. We'll get to that.

But first, let's talk about Mark Zuckerberg saying the quiet part out loud.

At an internal Meta meeting this week, Zuckerberg reportedly told staff that AI agents haven't progressed as quickly as he'd hoped. Now, this is a man who has, over the past eighteen months, made increasingly bold claims about agents taking over software engineering jobs by the end of this year. So when he turns around and says "actually, slower than we thought" — that's worth a moment.

Here's what he means, roughly. AI agents — these are AI systems that don't just answer questions but actually go and do things: book meetings, write code, navigate websites, complete multi-step tasks on your behalf — they're real, they exist, and they're improving. But the gap between "this can do a thing in a demo" and "I would trust this to do that thing inside my actual company without someone watching it" is still quite wide. The agents are getting better at individual steps. They're still surprisingly bad at chaining those steps reliably without going off-piste.

And here's the bit that matters for anyone making decisions about AI right now: the CEO of one of the most aggressive AI-investing companies in the world has essentially confirmed that the timeline for autonomous AI doing serious work is... fuzzier than the headlines suggested. Nobody really knows yet. Not even the people spending the most money on it.

We've covered before on this show how agentic AI is raising governance questions faster than anyone's answering them — who's responsible when an agent makes a decision that causes real harm? What does oversight even look like when the whole point is that it runs without permission check-ins? Zuckerberg being candid about the pace doesn't solve those questions. But it does suggest we've got slightly more time to ask them than some people implied. That's not a bad thing.

Meanwhile, on a completely different note — and I mean different — OpenAI has reportedly proposed giving five percent of its equity to a US sovereign wealth fund. Which sounds like a financial story, but is actually a political one.

The idea, as reported by the Financial Times, is that Sam Altman floated this as a way of easing tensions with the Trump administration and — and I'm quoting here — "blunting mounting public backlash against AI." The argument being: if the public has a financial stake in the gains from AI, they might feel better about the whole thing.

Now. There's a version of this that's genuinely interesting — the idea that AI wealth concentration is a problem and giving ordinary people a share of the upside is worth exploring. That's a real debate. But let's be clear about what's actually happening here. OpenAI is preparing to go public. They have filed their S-1 confidentially. They need Washington onside. And they've noticed that "the government should probably own a bit of us" is a useful thing to say when the government is deciding what rules to write about you.

We covered last week how OpenAI floated the idea of giving the US government a direct equity stake — this is the same story in a slightly different frame, where the sovereign wealth fund angle makes it sound more like public benefit and less like political deal-making. The distinction between those two things is, let's say, being worked quite hard.

What to watch: whether this actually makes it into the terms of any formal arrangement, or quietly evaporates once the IPO filing goes public and the lawyers get involved. Governments make for unusual shareholders. And when your regulator is also your investor, the dynamics get very interesting very fast.

Right. And finally — this one I want you to actually think about, because it moves quickly and the implications are bigger than the story itself might suggest.

Netflix is using an AI-generated version of Gene Wilder's voice in a new Willy Wonka reality show. Wilder died in 2016. The show premieres in September.

Now, I'll note upfront: I don't know the details of whether his estate consented to this, whether it's legally cleared, any of that — those details weren't available and I won't speculate. But the technical fact is straightforward: a streaming platform has used an AI voice clone of a deceased actor in a new commercial production, and they've put it in the trailer.

Here's why it matters beyond the "that's a bit creepy" reaction, which is valid. We are now in a world where the boundary between what a person created in their lifetime and what can be created in their name after they're gone is genuinely blurry. And it's going to get blurrier. The question isn't whether the technology can do this — it obviously can, you're listening to proof of that — it's who decides when it's acceptable. Estates? Audiences? Regulators? The streamer who has the rights?

There's a version of this that feels like a celebration of a beloved actor. There's another version that feels like the entertainment industry discovering that dead performers don't need contract negotiations. Hollywood has been worried about AI for a while — we covered the Oscar eligibility ban a while back — but this is a somewhat different category. That was about AI generating performances wholesale. This is about using a real person's voice, learned from their actual recordings, to say things they never said.

It won't be the last time we see this. I'd watch this space carefully if you work anywhere near media, entertainment, or intellectual property.

That's your lot for today. An honest CEO, a sandwich company doing its best to be a tech company, and Gene Wilder's voice making a comeback he didn't plan for. Three stories, a bit of politics, and at least one existential question about who owns a human voice. Not bad for a Thursday. See you next time.