When every company claims to be building AGI, AGI means nothing. Watch the industry find its next empty label.
For years, “AGI” was the most valuable word in AI.
OpenAI baked it into its founding mission. CEOs used it in earnings calls. VCs built funds around it. It was the North Star — the destination that justified every funding round, every hiring spree, every bold claim about what comes next.
Now watch them all quietly stop saying it.
The Verge recently documented a pattern: AI companies that used to talk about AGI constantly are finding new language. They haven’t changed their roadmaps. They’ve changed their vocabulary.
How a Word Gets Used Up
AGI started as an academic term with a real definition — artificial general intelligence, a system capable of learning and performing any intellectual task a human can. There were papers, debates, actual criteria.
Then the marketing teams got hold of it.
“We built a better chatbot” doesn’t raise billions. “We’re building AGI” does. So every new AI company adopted the same story. Anthropic said they were close. Google DeepMind hinted they were around the corner. xAI said they’d get there faster. Even startups that had barely trained a model were putting “Road to AGI” in their pitch decks.
When everyone claims to be arriving at AGI simultaneously, nobody is.
This is semantic inflation. It’s the same mechanism that destroyed “disruptive,” “paradigm shift,” and a dozen other tech buzzwords. The word doesn’t change — it just gets used so much that it stops carrying information.
The Exit Strategies Differ
The interesting part isn’t that companies are dropping AGI. It’s how differently they’re doing it.
OpenAI has the hardest pivot. This is the company whose constitutional mission is “ensure AGI benefits all of humanity.” Sam Altman’s public language has shifted noticeably — more “superintelligence,” more “frontier models,” more specific capability claims, fewer AGI declarations. The reason is obvious: when you promise AGI and deliver increasingly capable chatbots, the gap becomes a liability.
Anthropic played this differently from the start. They barely ever said AGI. Their framing was “constitutional AI” and “helpful, harmless, honest” — product language, not prophecy language. In an environment where AGI claims got progressively more ridiculous, Anthropic gained credibility by not playing.
A third group went metric-driven. Instead of claiming AGI progress, they publish benchmark scores. It’s more honest, more verifiable, and significantly less exciting. Nobody clicks on “Model achieves 87.3% on MMLU” the way they click on “We’re approaching AGI.”
Three Signals Behind the Shift
This isn’t just marketing rotation. Three structural forces are pushing AGI off the stage.
Capital is asking for numbers now. For years, “we’re building AGI” was a funding thesis. That era is over. Investors want revenue, moats, paths to profitability. AGI went from an asset on the balance sheet to a liability — because every month that passes without AGI arriving, the gap between promise and product gets wider.
Regulation is catching up. Declaring that you’re building AGI is basically a regulatory invitation. Europe already has legislation targeting frontier AI models. US Congressional hearings are growing more frequent and more pointed. In this environment, saying “we’re building AGI” is like putting up a sign that says “inspect us first.”
The talent game changed. The best AI researchers — the people who actually push the frontier — are immune to AGI slogans. What attracts them now is concrete technical problems, interesting datasets, and open research environments. “Join us to achieve AGI” works on MBA students, not on people who read the latest arXiv papers.
What Comes Next
The AI industry won’t stop creating labels. It’s too good at packaging technology as destiny.
“Superintelligence” is OpenAI’s preferred replacement — it sounds more concrete than AGI while keeping the grand narrative intact. “Frontier AI” is the term regulators and academics have converged on. “Autonomous agents” is the capability-focused narrative that shifts attention from a final destination to incremental progress.
All of these will follow the same arc.
The lifecycle of an AI industry buzzword is roughly two to three years: a new term emerges, a few early adopters use it and gain credibility, then every company piles on, the meaning inflates, and eventually it becomes dead weight that the next wave of smart operators abandons.
AGI completed this cycle. The next word will too.
The real question isn’t what label AI companies will use next. It’s whether the industry ever learns to talk about what it’s actually building instead of what it’s selling.