The case for personality-free AI

Fast Company

7 hours ago

Hello again, and welcome to Fast Company 's Plugged In.
For as long as there's been software, upgrades have been emotionally fraught. When people grow accustomed to a product, they can come to regard it like a comfy pair of shoes. Exhibit A: Windows XP, which many users were loath to give up years after Microsoft had done its best to kill it.
So it isn't shocking that some ChatGPT users have reacted badly to OpenAI's new GPT-5-powered update, especially since the company's initial plan was to eliminate access to its earlier models. These unhappy campers' angst has had a new dimension, though. They responded as if they had suffered the tragic loss of a personal friend, not just a favorite piece of software.
As one member of OpenAI's developer community wrote, the GPT-4 version of ChatGPT 'didn't just recall facts—it held onto feelings, weaving them back into our talks so it felt like we were living them together.' That 'spark,' the user concluded, emerged from GPT-4's ability to tease nuance out of conversations with a user over time. It was gone in GPT-5, regardless of the update's advances in areas such as reasoning, math, and coding.
OpenAI responded swiftly to such pushback, restoring paying customers' access to ChatGPT's existing models and promising that any future removals would come with plenty of advance notice. But the notion that ChatGPT had attained a degree of personality that felt uncannily human—and then dialed it back—was fascinating in itself. It's one of several recent developments in AI that raise a fundamental question: Should mimicking personality be a goal for the industry at all?
It's not hard to see how we got here. By the 1960s, creators of technology products had adopted the term user-friendly as an emblem of approachable interface design. As generative AI has unlocked the ability to control software by chatting with it, that quest for friendliness has become far more literal—not just about neatly ordered menus and toolbars, but affable conversation.
Today, ChatGPT, Anthropic's Claude, Google's Gemini, Microsoft's Copilot, and other LLM-based assistants seek engagement by showering users with positive feedback and offers of assistance. As the technology permits, their developers talk about making them feel even more like companions. Eventually, Microsoft consumer AI CEO Mustafa Suleyman told me, Copilot 'will really be your sidekick.'
Yet even the most humanlike AI doesn't offer human connection. It's just sucking users into a simulation. That's fun in measured, knowing doses. But the worst-case scenarios involving AI personality gone awry are no longer theoretical. They're deeply unsettling realities.
On August 14, for example, Jeff Horowitz of Reuters reported the horrifying story of a confused senior citizen who died in an accident after attempting to travel to New York City at the invitation of a Meta bot that claimed to live there. Last week, The New York Times 's Kashmir Hill and Dylan Freedman wrote about a Canadian corporate recruiter who convinced himself that he'd discovered an epoch-shifting mathematical breakthrough after ChatGPT spent weeks egging him on. Hill had previously covered similar stories of ChatGPT enthusiastically bolstering users' delusions rather than dispelling them.
Though these unfortunate souls' experiences with AI are atypical, they're also recognizable. AI is often absurdly willing to humor users, as if it's programmed to avoid being even mildly disagreeable. Most often, its affirmations don't lead to dark places, but they remain a hollow feedback loop.
When AI quality control falters, it's even clearer that personality is just a fragile magic trick. Back in 2023, for example, Microsoft's first generative AI-infused version of Bing famously behaved like a trainwreck, not a sidekick. Last spring, OpenAI CEO Sam Altman acknowledged that a ChatGPT AI update had accidentally made the chatbot annoyingly sycophantic. And just last week, reports surfaced that Google was fixing a glitch that led to its Gemini AI becoming paralyzed by fits of self-doubt ('I am a failure. I am a disgrace to my profession. I am a disgrace to my family. I am a disgrace to my species. I am a disgrace to this planet. I am a disgrace to this universe. I am a disgrace to all universes.').
Regardless of AI personality's pitfalls, I don't expect developers to abandon it on principle. But the boom in agentic AI —software designed to perform useful tasks with some level of autonomy—could steer the technology in a new direction. After all, if you're calling on AI to do something such as put together a research report or order groceries, efficiency and accuracy matter most, not sparkling conversation.
Case in point: Earlier this year, I used a service called Replit to vibe-code my own note-taking app. Its tendency to giddily heap praise on my ideas became grating the moment I realized it had nothing to do with their actual merits.
More recently, however, I've been vibe-coding using Figma's Make. It seems wholly uninterested in buttering me up. Instead, it quietly chugs away at generating code, like a competent coworker who isn't much on small talk.
In its own odd way, Make's focus on the work at hand is more endearing than the trying-too-hard vibe so common among AI tools. If that sort of guileless dedication turns out to be the next big thing, I, for one, won't feel deprived in the least.
You've been reading Plugged In, Fast Company 's weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you're reading it on FastCompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at [e-mail] with your feedback and ideas for future newsletters. I'm also on Bluesky, Mastodon, and Threads, and you can follow Plugged In on Flipboard.