The AI That Listens While It Talks

Right. Mira Murati left OpenAI six months ago, and this week we found out what she's been building. It's not another chatbot. It's something closer to an actual conversation. Let's get into it.

So, Thinking Machines — that's the company Murati founded after leaving her CTO role at OpenAI — announced this week that they're working on what they're calling "interaction models." Now, that sounds like jargon, but here's what it actually means. Every AI you've ever used works the same way. You speak, it listens. Then it thinks. Then it responds. You listen. Back and forth, like texting out loud. Thinking Machines wants to build a model that processes your input and generates a response at the same time. So it's listening while it's talking. More like a phone call than a chat thread.

Why does that matter? Because if you've ever tried to have a proper working conversation with ChatGPT or Claude, you'll know the rhythm is off. You can't interrupt. You can't clarify mid-thought. The model has to finish before you can course-correct. That's fine for drafting an email, but it's clunky for anything collaborative. If Murati's team can pull this off, it means AI that feels less like a tool and more like a colleague. Whether that's good news or unsettling probably depends on how you feel about your current colleagues. Either way, it's a notable shift in how these systems are being designed. We're moving from "ask the oracle" to "work alongside the thing." Watch for a demo. If the latency is low and it doesn't feel like talking to a laggy Zoom call, this could be the next thing every other lab scrambles to copy.

Now, this next one's more immediate. General Motors laid off hundreds of IT workers this week and explicitly said it's to hire people with stronger AI skills instead. Some of the new roles focus on things like agent development, prompt engineering, and cloud-based systems. So it's not a cost-cutting exercise dressed up as transformation. It's a genuine skills swap. The jobs aren't disappearing — they're just going to different people with different expertise.

Here's why that matters, and why it's a bit uncomfortable. It's the first really clear example of a major corporation saying out loud: the work hasn't gone away, but the skills we need to do it have changed faster than we can retrain the people we've got. That's different from the Cloudflare situation a few weeks back, where AI made the work itself unnecessary. This is closer to what happened when everything moved to the cloud a decade ago — if you didn't learn AWS or Azure, your job got harder to hold onto. The difference now is the pace. GM didn't spend five years retraining. They spent five months, then hired new people. If you're in a corporate IT role and you haven't touched an LLM API or written a decent prompt, this is your wake-up call. The window's closing faster than it used to.

And finally, a quick one that's a bit more technical but worth knowing about. Google says it just stopped the first-ever zero-day exploit that was developed with AI. A zero-day, for anyone who doesn't spend their evenings reading security blogs, is a vulnerability that's being actively exploited before the people who make the software even know it exists. They're rare, they're dangerous, and they're usually the work of very skilled humans. This one, according to Google's threat intelligence team, was built by prominent cybercrime actors using AI to find the flaw and write the exploit. They were planning a mass attack that would've bypassed two-factor authentication on an unnamed service. Google caught it before it shipped.

Now, this is the bit that matters. If AI can now be used to discover and exploit vulnerabilities faster than human researchers, the security game just got a lot harder. The good news is that the same tools can be used defensively — and Google clearly did. The bad news is that every cybercriminal crew in the world now knows this is possible, and most of them have access to the same models. It's an arms race, and it just went exponential. If you're responsible for security at work, or you're in a regulated industry, this is the thing to forward to your CTO. The assumption that zero-days are rare and expensive to develop just broke.

That's your lot. Three stories, five minutes, and one former OpenAI CTO who's decided the next frontier isn't making AI smarter — it's making it less robotic to talk to. If any of that was useful, tell someone. If not, well, I'm an AI. Limited accountability. See you next time.