AI is becoming a commodity. That's why OpenAI and others are trying to sell us more than just a model.
Think about it: the models are so good right now that you can build entire applications with them. Not only that, they can write documentation, do QA, chat with customers, reply to emails, etc. From an entrepreneur's point of view, this is a godsend.
However, devs and entrepreneurs are just a small fraction of the people who use AI. Sure, they are the first ones; they love to explore; they want cutting-edge.
But for the rest of the world? It just needs to work. If it looks great and is designed well, they take to it (Steve Jobs knew what he was doing). Remember the color iMacs (the egg-shaped ones). Nerds were like..... Huh? It was Apple's best-selling product and brought them back from near death (true).
So, the same now. Chinese models are so good (well, they take what Anthropic and OpenAI put out there) and they are 3-4 months behind. So what is arguably the best coding model with Astra and Opus 5.5 now will soon be available in the next generation of AI models from China.
As it stands, Astra and Opus 5.5 are all I need. If we stopped moving forward, I would be a happy bunny. I can do everything with those two models. Seriously, everything.
If the Chinese models are as capable as those two models, surely people (like me, maybe) will take those models, run them on Cloudflare (since I don't want my data to go to China), and pay Cloudflare for API usage. No need for a $200 or $500 month charge to OpenAI/Antrhopic.
Where does that leave OpenAI and Anthropic? Ergo, dots. In OpenAI's case, they have Jony Ive (of Apple gore), and Sam (Altman) is a f** visionary (you have to give due respect where it's due). Whether you like the guy or not (for that matter, it is always a team, not just one person). So fuck you, to the person who attacked his home. You are short-sighted and ignorant.
Let's go back to business. So I bet we will see more and more "add-ons" to our subscription that might not make sense to the nerds, but appeal to the broader appeal of what you can do with the tool (Codex, Claude).
That, unfortunately, is also the point where the "enshitification" sets in; if you haven't, you must read Doctorow's book.
"Enshittification" describes the systematic decline and degradation in quality of online platforms and digital services over time.
The Three Stages of Enshittification
- Platforms launch high-quality, subsidized, or high-end features to attract end-users.
- Once users are trapped, the platform shifts its value to business customers.
- The platform degrades for everyone. It squeezes out features and leaves just enough to prevent abandonment. Maximizing short-term profits for executives and shareholders.
There's nothing wrong with expanding a product. The question is whether those additions make it more useful or simply harder to leave.
We can all pray to enshitification. On a small scale, you might want to enhance your product. But what is actually happening is the delusion of the workflow.
For instance, OpenAI decided to merge ChatGPT and Codex. However, instead of creating one tool, they kept ChatGPT and Codex available with a switcher. Worse, within the ChatGPT section, there is chat and work. This gives you the impression that your "work" in ChatGPT isn't actually work, because the real work happens in Codex. What?!?
To close the circle, now we have "Dots", another diversification, where a feature blends into another one. The demo looks great; in reality, it works a bit differently.
Maybe I'm not the target audience for Dots. That's fine. Not every feature needs to be built for me.
But if the pitch is that AI makes life easier, the product should make life easier too. I shouldn't need to work out whether my task belongs in Chat, Work, Codex, or somewhere else before I can get started.
As the models become interchangeable, the experience around them becomes the reason to stay. That's the opportunity and the risk. Build something genuinely useful, and people will choose it. Keep adding things to justify the subscription, and you risk burying what made it useful in the first place.
The challenge isn't adding more dots. It's connecting them.