This is Scale & Strategy, the newsletter that's like having a Bloomberg terminal, but with personality.
Here’s what we’ve got for you today:
Jensen Huang says stop chasing “AI-proof” careers
ChatGPT doesn’t have one personality anymore
Jensen Huang says stop chasing “AI-proof” careers
Jensen Huang is pushing back on the growing obsession with “safe” degrees and AI-proof career paths, arguing that the better question isn’t what survives AI, but what gets amplified by it.
Speaking with CNA, Huang said students shouldn’t optimize around fear-based subject selection. Instead of asking which fields AI can’t touch, he suggested asking how AI can elevate your learning, craft, and effectiveness inside the thing you actually care about doing.
That sounds obvious until you realize half the internet is currently treating career planning like people are hiding from an incoming asteroid.
Huang used journalism as an example. Good reporters don’t just generate questions. They listen, adapt, read context, understand audience psychology, and know when something unsaid matters more than the prepared script. AI can assist pieces of that workflow, but it doesn’t automatically produce judgment.
That distinction keeps getting lost in the broader AI panic cycle.
He also referenced “wabi-sabi,” the Japanese idea that imperfection itself carries value, arguing that distinctly human qualities may become more important, not less, as AI systems get more capable.
There’s truth there, though people tend to romanticize it too quickly. Taste, judgment, timing, social intuition, narrative framing, emotional calibration, these things matter more in AI-heavy environments precisely because raw execution is becoming cheaper.
The weak take is “AI replaces humans.”
The equally weak take is “human creativity will magically save everyone.”
Reality is less comforting and more uneven. AI compresses the value of average output while increasing the value of high-leverage decision-making, direction-setting, and synthesis. The middle gets squeezed first.
Huang also called the current narrative around AI-driven layoffs “lazy,” pointing out that AI has barely arrived at scale and is already being blamed for workforce reductions everywhere.
He’s partially right. A lot of companies are using AI as a convenient umbrella explanation for restructurings they probably wanted to do anyway. “AI transformation” sounds cleaner on earnings calls than “we overhired during zero-rate euphoria and now need margin discipline.”
At the same time, pretending AI isn’t accelerating labor compression in certain functions is becoming hard to defend. More than 80,000 jobs have already been cut across major companies this year while AI spending ramps aggressively upward. Those two lines on the chart are not unrelated no matter how much PR departments try to blur them together.
The more useful takeaway is probably this: trying to predict “safe” subjects is a losing game because the tooling is evolving too fast for static career maps to hold.
The people who benefit most from AI usually aren’t the ones avoiding it. They’re the ones building taste, judgment, domain depth, and enough adaptability to use the systems better than everyone else in their field.
Which is less comforting than “learn to code” and less poetic than “humanity will prevail,” but unfortunately much closer to how labor markets actually work.
Framer helps teams design, build, and launch their marketing sites lightning fast. With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster. Trusted by companies like Miro, Bilt, and Perplexity
Speed without chaos: ship pages and updates faster without turning the site into a fragile set of one-off hacks
Reduce dependency: shift routine brand and marketing work out of product engineering queues.
Production-grade foundation: Run real marketing systems (CMS, SEO, performance optimization) with governance and collaboration
Most people still think of ChatGPT as a single product with a single voice. In reality, OpenAI is increasingly turning it into a configurable interface layer where tone, behavior, memory, and interaction style can shift depending on the user.
That matters more than it sounds.
What started as basic “Custom Instructions” back in 2023 has evolved into a much deeper personalization system. Users can now shape how ChatGPT responds through preset personality traits like Friendly, Candid, or Cynical, alongside Memory settings that store preferences, context, occupations, nicknames, and custom behavioral instructions across conversations.
Underneath all of this is a pretty important strategic shift: OpenAI is moving from building “a chatbot” to building adaptive conversational infrastructure.
In an interview with The Deep View, OpenAI product manager Laurentia Romaniuk explained that personality preferences are increasingly treated as part of user context. The idea is simple: a brainstorming partner shouldn’t sound like a legal compliance bot, and a coding assistant probably shouldn’t talk like a mindfulness coach who just discovered herbal tea.
Users are also getting surprisingly specific with instructions. Some configure writing styles down to sentence rhythm and tone calibration. Others try to replicate their own communication patterns almost exactly. Which is either fascinating or the first stage of everyone outsourcing their personality into reusable presets. Hard to tell yet.
The important technical point is that personalization does not create a separate model for every user. The underlying system remains the same. What changes is the behavioral layer wrapped around it.
That distinction matters because people often confuse tone with capability. A warmer model can feel smarter even when it isn’t. A concise model can feel colder even when it’s more useful. Small wording shifts radically change how humans interpret intelligence, competence, empathy, and trustworthiness. Humans are extremely susceptible to interface psychology. Entire industries exist because people trust nicer fonts.
Romaniuk said OpenAI’s goal is to make the default experience broadly useful while allowing deeper adaptation over time. The hard part is balancing warmth, clarity, naturalness, and restraint without drifting into something overly flattering, emotionally manipulative, or weirdly synthetic.
That challenge is becoming increasingly important because chatbot personality is no longer just UX polish. It directly shapes user attachment, trust, behavior, and perceived reliability.
You can already see this playing out across the industry. Platforms like Character.AI exploded largely because users wanted specific personalities, not just raw capability. OpenAI itself got backlash after retiring versions of GPT-4o that many users preferred specifically because of their tone and conversational style.
And this is where things get more complicated than “users want customization.”
The more conversational and adaptive these systems become, the more people naturally anthropomorphize them. That creates real product advantages, but also real risks. A chatbot that feels emotionally intelligent can influence users far more effectively than one that feels obviously mechanical, even when the underlying system is fundamentally the same prediction engine underneath.
OpenAI says its goal is to balance personalization with safeguards and user agency, which is directionally correct. But the broader issue probably isn’t solvable by one company tweaking system prompts and safety layers.
Once AI systems become persistent, adaptive, emotionally calibrated, and integrated into daily life for hundreds of millions of people, personality itself becomes part of the product surface area. At that point, the conversation stops being just about model capability and starts becoming about behavioral influence at scale.
Which, in typical tech industry fashion, means we accidentally turned “how should the chatbot sound?” into a civilization-scale UX problem.
Framer helps teams design, build, and launch their marketing sites lightning fast. With the ability to publish hundreds of CMS pages in a single click, operate at a global scale with seamless localization, and even host unified content across multiple domains, teams have never been able to ship faster. Trusted by companies like Miro, Bilt, and Perplexity
Speed without chaos: ship pages and updates faster without turning the site into a fragile set of one-off hacks
Reduce dependency: shift routine brand and marketing work out of product engineering queues.
Production-grade foundation: Run real marketing systems (CMS, SEO, performance optimization) with governance and collaboration
