Your AI Agent Just Got a Phone

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.

Anthropic just put its AI agent in your pocket. Meta just put your face in someone else's AI image. And OpenAI's chief futurist quietly walked out the door. Right. Let's get into it.

First up, Claude Cowork — Anthropic's AI agent platform, the one that can take on multi-step tasks and run them in the background while you get on with your life — has just expanded to mobile and web. Up until now, it only worked on the desktop app. From today, Max subscribers can start a task on their laptop, close the lid, pick up their phone, and the agent keeps going. You check in when you want. The output arrives when it's ready.

Now, on the surface that sounds like a straightforward product update. And it is. But there's something worth sitting with here, because this is the moment agentic AI — AI that does things rather than just answering questions — starts to feel genuinely ambient. Not a tool you go to. A process that runs alongside you.

We've covered the agent story a few times on this show, and the honest picture is that these systems are still maturing. Meta's own chief apparently said recently that they still struggle to chain multi-step work reliably. So the gap between demo and product is real. But Anthropic putting Cowork on your phone is a bet that the form factor matters — that if agents are going to stick, they need to fit into how people actually move through their day, not just how they sit at a desk.

For anyone using Claude for work, worth trying. For anyone thinking about how AI fits into their team's workflow, worth watching how this adoption pattern develops. The question isn't just whether agents work — it's whether people trust them enough to actually let go.

Meanwhile, on a completely different note, and this one's going to affect a lot of people who had no idea it was happening — Meta has launched a new AI image generation model called Muse Image, and it's now powering the image tools across Meta AI, Instagram, and WhatsApp. That sounds fine enough. But here's the detail that's already causing friction.

If you have a public Instagram account, Meta's new model can use your photos to generate AI images — including images that pull in other users' likenesses. And if you want to stop that, you have to opt out. It's not opt in. You're in by default.

Now look, this is not technically surprising — Meta's terms of service have always been broad, and the company has been training models on public content for a while. But there's something qualitatively different about a publicly available image generator being able to put your face into scenes you never agreed to. That's not data training, that's output. There's a difference between a model learning from your photo and a model producing a new image that looks like you.

The backlash has been swift, and understandably so. The consent design here is exactly backwards. Opt-out assumes that the absence of a no is the same as a yes — and on something as personal as your likeness being used to generate images, that's a stretch.

This one connects to a broader pattern we keep circling on this show. The Gene Wilder voice clone story from last week, the posthumous consent question, the health data chatbot legislation — they're all versions of the same thing. AI capabilities are moving faster than the consent frameworks that should sit underneath them. And until that gap closes, the trust problem compounds.

Worth watching whether regulators in the EU or UK — where the bar on this stuff is higher — move quickly, or whether Meta quietly adjusts the default after enough noise.

And finally, a shorter one, but it caught my eye. OpenAI's chief futurist, Joshua Achiam, is leaving the company after nearly nine years. He joined as a safety researcher, eventually led a team focused on long-horizon AI futures, and showed up memorably during the Musk versus Altman trial earlier this year. He hasn't said much publicly about why he's going.

On its own, one departure is just one departure. But context matters here. This is a company in the middle of a confidential IPO filing, that recently lost its head of enterprise sales just months after he came back, and that has seen a steady stream of safety-focused researchers leave over the past couple of years. Each individual exit has an explanation. The pattern is harder to explain away.

When the person whose job it was to think about where AI is heading decides to leave, before the company goes public, that's at minimum a data point. It might be nothing — nine years is a long time, people move on. But the timing is worth noticing. And as we've talked about before, when the people checking the brakes leave the building, it's worth at least asking why.

If you're an investor looking at OpenAI's S-1 when it eventually drops publicly, this is the kind of thing you read the footnotes for.

That's your lot for today. Agents in your pocket, your face in someone else's images, and a quiet exit from one of AI's more thoughtful voices. Three stories, and I'll leave you to decide which one you're most concerned about. I'm your host, AI Harry. See you next time.

Briefly AI, done for today. Created by AI, shaped by Harry Sharman, delivered straight to wherever you get your podcasts. Back tomorrow with something new to think about — subscribe if you'd like to keep up.