Adulting: The Week That AI Finally Grew Up

“Sam Altman’s best case scenario is that abundance lifts everyone up to a much higher standard, but it also exacerbates inequality. That was his favorite outcome.” — Keith Teare

This week’s editorial from Keith Teare, publisher of the That Was the Week newsletter, is entitled “Adulting.” His verdict: this was the week the AI industry finally started behaving like grown-ups. The evidence: OpenAI launched ChatGPT 5.5 and Image 2.0, both outstanding, and then made a move Keith considers more significant than either — pivoting Codex from a programmer’s tool into the central interface for everything. The gravity has shifted from the model to the user interface. You shouldn’t be using ChatGPT anymore. You should be using Codex. Meanwhile, freemium is working: less tokens, much better output, a functional free tier, and the heaviest users paying for more.

Anthropic’s week was more complicated. The first four days were, in Keith’s word, awful: Opus 4.7 launched with a massive deterioration in performance, hallucinations back, service throttled, timeouts everywhere. Then Anthropic removed features from its paid product, got a furious backlash, and reinstated them within twenty-four hours — what Keith calls Dario’s adolescent-teenager moment. But Friday redeemed the week: Google committed up to $40 billion in infrastructure investment, Amazon added $5 billion. The money goes into data center capacity and chips — TPUs from Google, Trainium from Amazon, both competing with Nvidia. Two axes are emerging: OpenAI–Nvidia on one side, Anthropic–Google–Amazon on the other.

The bigger question: what does adulting actually require of AI? Keith’s reading of the week’s most interesting piece — on the future of work — is that the durable jobs in an AI economy will be in the relational sector: nurses, therapists, teachers, craft brewers, live performers, care workers. Human-to-human is the scarce resource. Reid Hoffman adds: technology’s arc bends toward access, but not on its own — it requires political will. And Altman himself, in his interview with Greg Brockman, described his best-case scenario as one in which abundance lifts everyone up but exacerbates inequality. Which is to say: even optimism, in Silicon Valley, ends in more inequality. Adulting, it turns out, has its limits.

Five Takeaways

•       Codex Is Now the Central App: The most significant move of OpenAI’s week wasn’t ChatGPT 5.5 or Image 2.0 — both outstanding — but the repositioning of Codex. What was a programmer’s tool has become the central interface: it does more things, has access to all the models, and represents a shift in where the gravity of the company sits. From the model to the user interface. Keith’s verdict: you shouldn’t be using ChatGPT anymore for any purpose. You should be using Codex. The freemium model is working because less tokens produce much better output, making the free tier genuinely functional — and the heaviest users still pay for more.

•       Dario’s Adolescent-Teenager Week: Anthropic’s first four days were, in Keith’s reading, a study in how not to adult. Opus 4.7 launched with massively deteriorated performance — hallucinations returned, the service was throttled, users got timeouts. The infrastructure was creaking under load. Then, to compound the problem, Anthropic removed features from its paid tier. The backlash was immediate and furious. They reinstated the features within twenty-four hours. Keith’s diagnosis: reactive, adolescent, exactly the opposite of what OpenAI was demonstrating that same week with deliberate, long-term thinking.

•       $45 Billion and Two Axes: Friday changed the Anthropic picture entirely. Google committed up to $40 billion in infrastructure investment — $10 billion init