That’s why AI commerce built around answers is so appealing. A clean, truthful result feels almost revolutionary at this point. I just hope AI chats from frontier labs—OpenAI, Anthrophic—don't t fall into the same trap.

It’s wild how normalized that’s become. We treat the noise as discovery, when it’s really just revenue choreography. A search bar that pretends to help but mostly sells.

Are games already a failed experiment for Netflix in your mind?

Yeah I guess, too, that OpenAI is the only company investing across the entire stack. They realized you can’t just pretrain your way to AGI anymore. You also need inference infrastructure, product infrastructure, deployment infrastructure. These all need to advance in parallel.

fwiw, I think Altman’s recent interview with Ben Thompson is full of hints about this:
- “Apps in ChatGPT” - getting more value from existing models
- “Instant Checkout” - better integration, not better AI
- Infrastructure deals - preparing for inference scale, not just training

So your read is OpenAI knows scaling is dead (or slowing), pivoted to inference/deployment as the strategy, but is publicly maintaining the scaling narrative to justify the infrastructure deals and keep the market hype going.

Is that roughly your thesis?​​​​​​​​​​​​​​​​

Love the .fyi

Imagine being a pro athlete and your weight loss makes national news.

Totally agree that you don’t want runaway costs. I guess I’m just saying what comes back isn’t necessarily methodologically sound. Cloudflare’s conflating “AI that can buy and summarize sources” with “AI that can do research.” But I have hope future models will be far more sophisticated.

3/ The real power here isn’t giving an agent $50 to “buy research.” It’s an AI that can evaluate whether a paywalled paper is worth purchasing based on how it advances the current investigation. That’s actual research methodology, not content shopping.​​​​​​​​​​​​​​​​

2/ The “deep research with a budget” example feels off though. Real research isn’t shopping with spending limits, it’s investigative work that follows citation trails and adapts based on findings. An AI should justify paying for sources because they fill research gaps, not fit budgets.