Does AI have a zone of genius?

The concept of “zone of genius” isn’t new. In my opinion, it’s been overused and generally overpromising. But it’s made a comeback. Now in the context of AI - find your “zone of genius” before AI takes your job, specifically finding that part that AI still can’t replace. Genius, right? But what could that be? It's the contextual stuff - reading the room you're in, hearing what's unsaid, making judgment calls that depend on a decade of lived experience in a specific domain. The practical version: if a task feels like filling in a form, AI can probably do it. If it requires knowing the people, the history, and the stakes - that's yours. It’s not something to put on your CV but it gets you brownie points during the next round of cuts or restructuring.

In other words - AI is like the GPS that takes you via the shortest and quickest road, even if that means dirt roads and dangerous intersections; while someone living there would avoid because they know better. Be that local guy/gal who simply knows what the GPS doesn’t.

For the past 18 months, the dominant corporate logic was: fewer people, more AI. Turns out it’s not that simple. Companies are hiring again - not despite AI, but because of it. How? Because AI agents need people to work alongside them, not instead of them. In the same numbers? I doubt it. Junior roles are back in demand because entry-level workers are “AI-native, not calcified in thought” (and cheaper…). Having coding agents doesn't mean engineers aren’t needed anymore. But here’s the plot twist we saw coming - the gains won’t be evenly distributed. Disruption is here for sure, but the timeline and outcome are more complicated than the headlines suggested.

Young adults now use ChatGPT not just for texts and dating app openers, but for real-time talking points during face-to-face conversations. Pulling up AI summaries mid-conversation to sound like they’ve read things they haven't. The clinicians are concerned because using a bot to pre-polish everything “slowly teaches you that your own read can't be trusted.” For people with social anxiety, the relief is real but temporary, and the tool can become a crutch.

I’ve had something similar happen in real life - while talking to someone about a famous doctor and her claims, the person I was talking to pulled out his phone and asked AI about what I just said. It was a breach of perfectly pleasant conversation that I was now looking to exit in under three seconds. Veto.

Kimi, DeepSeek, and a handful of other Chinese-built models are gaining real traction - cheaper, often open-source, and increasingly competitive on capability. They're making inroads despite the geopolitical headwinds. The pattern is familiar: a dominant player gets complacent, a cheaper alternative closes the gap, and switching is easy. This is already happening at the developer layer. Whether it reaches mainstream consumers depends on how long the price and performance gap holds.

ChatGPT Health is rolling out broadly - a feature that helps you track symptoms, understand lab results, and navigate health decisions. The problems are like a list of symptoms a hypochondriac would have: long. No access to your actual medical history, no accountability if it gets something wrong, and a tendency to sound more confident than it should. Point in case: my son had several blisters that looked worrying, and despite telling AI that we were at the seaside and other vivid details, AI suggested this is a big deal and we should get medical attention. If I wasn’t a non-panicky mom, I would have run to the hospital. Turns out, they were mosquito bites and disappeared within 24h. Maybe it can also save lives, but proceed with caution. Research yes, diagnosis no.

Google has added selfie video as an account sign-in option - a short recorded clip instead of (or alongside) a password or 2FA code. It's framed as a security improvement, using liveness detection to block static photo spoofs. The obvious tension: you are now training Google's biometric systems with your face as a condition of account access. As if we didn’t already look silly lifting the phone up to take a photo of ourselves, now we have to do it at work, too?

A professor embedded white-on-white instructions in his exam telling ChatGPT to produce wrong answers. Students who fed the whole document into an AI got those wrong answers back - and submitted them. What an elegant trap. The students weren't just using AI; they were using it without reading the question first, nor their answer before submitting it. If these students are the future lawyers, mechanics or doctors - we are all going to hell in a handbasket. Well done, professors, for teaching these kids a lesson. Pun intended.

Know someone who would find this information useful, nay, indispensable? Forward the email to them! If someone forwarded this to you (aww, how nice of them to think of you), subscribe at tooproductive.com.

Till next time,

Jana