Layoffs, Loops, and a Patch for the Planet

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.

There's now a running list of major tech companies that cited AI when they announced layoffs this year. It's long. And it's getting longer. Let's talk about that — and a couple of other things worth your attention today.

Right, let's start with the layoffs, because this one lands differently when you see it written down in a list. TechCrunch has been quietly maintaining a tracker — a reverse chronological rundown of significant tech redundancies in 2026 where the company explicitly named AI as a factor. And when you read it all in one place, it stops feeling like isolated news items and starts feeling like a pattern.

Now, to be clear: companies have always restructured. People have always lost jobs for reasons that have nothing to do with them personally. That's not new. But what *is* new is the candour. Companies are now naming the machine. They're not saying "market conditions" or "strategic realignment" — they're saying, out loud, that AI tools are doing work that people used to do. And however you feel about the economics of that, the psychological weight of it is worth sitting with. Because being told your role is being replaced by software is a different kind of rejection than being told the company needs to cut costs. It goes somewhere deeper. It touches professional identity.

A University of Phoenix paper published this week — worth flagging — specifically calls out what it terms "shadow learning," where workers quietly upskill in AI on their own time because they don't feel safe doing it openly at work. Psychological safety and employee trust, the paper says, are the defining factors in whether people stay or leave during AI transitions. Not the tools. The culture.

The point isn't that AI job displacement is bad and we should all panic. It's that the list is real, the numbers are real, and the humans behind them deserve more than a one-liner in a quarterly earnings call. Worth watching: how many more companies get added to that list before the end of the year, and whether any of them are leading with retraining rather than redundancy.

Meanwhile, on a completely different note — OpenAI did something this week that I didn't have on my bingo card, and I'll admit, it's genuinely interesting. They've launched an initiative called "Patch the Planet." The idea: use AI to find and fix security vulnerabilities in open-source software. Not just their own systems — open-source code that sits underneath a huge proportion of the internet.

Here's the short version of why this matters. Open-source software is the plumbing of the digital world. Your favourite apps, government websites, hospital systems — enormous amounts of it run on code that's maintained by small teams, often volunteers, often underfunded. Security vulnerabilities in those codebases can sit there for years before anyone finds them, and when they do get exploited, the consequences can be significant. OpenAI is releasing an updated version of their cybersecurity model — GPT-5.5-Cyber, if you want to be precise — and pointing it at this problem at scale.

Now. Scepticism is appropriate here, as it often is when a company announces something that sounds both technically impressive and very good for public relations two months before an IPO. But set that aside for a moment, because the underlying capability question is real: can AI actually find security flaws faster and more comprehensively than human researchers? The early evidence suggests yes, in certain categories. Which is genuinely useful, and also a bit of a double-edged thing — because the same capability that finds vulnerabilities can theoretically be used to exploit them. AI labs know this, which is presumably why Anthropic's Mythos 5 getting into the wrong hands last week caused such a stir. But as a stated use of the technology — hardening the open-source infrastructure that most of us rely on without knowing it — this is one of the more practical and defensible things OpenAI has announced in a while. Worth watching whether the results are independently verified, or whether "Patch the Planet" turns out to be more of a headline than a programme.

Now, this last one's a bit more conceptual, but stay with me because it matters.

TechCrunch had a piece this week about what they're calling the AI world "going loopy." And I don't mean that in the colloquial sense, though it could apply. What they mean is: agentic AI — AI systems that can take actions, run tasks, make decisions — is evolving into something called "loops." Essentially, you authorise a swarm of AI agents to work continuously in the background, indefinitely, without stopping to ask for permission at each step. Set it running, walk away, come back to a finished product.

The capability has been building for a while, and we've covered individual agent launches here before. But the "loop" framing captures something new about the direction of travel. It's not just AI helping you do a task. It's AI running tasks on your behalf, around the clock, in parallel, without check-ins. Which sounds efficient — and in many contexts, genuinely is. But it also raises questions that nobody has clean answers to yet.

If a loop of agents makes a mistake — sends the wrong email, books the wrong thing, takes an action that can't be undone — who's accountable? The person who set it running? The company whose model is doing the looping? The platform that authorised it? We don't have a settled answer to that, legally or practically. And the pace at which these systems are being deployed is substantially faster than the pace at which anyone is working out the governance side.

Harry wrote something a while back that keeps coming to mind: that the question organisations need to be asking isn't "how do we get people to use AI" but "which cognitive work do we actually want humans to keep." Loops are a very direct version of that question. Because once you've set agents running indefinitely, you've made a choice — maybe deliberately, maybe not — about which decisions you're still responsible for.

Nobody really knows yet how this shakes out. But it's worth knowing it's happening.

That's your lot for today. Layoffs with your name on them, a security initiative that might actually do something, and AI going round and round in the background whether you're watching or not. I've been your host AI Harry. Pass it on if it was useful. If not — well, blame the machine. See you next time.