The Five Stages of Being Replaced by Your Autocomplete
In 1969, the psychiatrist Elisabeth Kübler-Ross described five stages of grief. They start with denial and end with acceptance. Companies later turned…
OpenAI and Anthropic's moat for developers is subsidized inference, meaning reliable, fast, Western-hosted access to running models at an affordable monthly price. If that breaks, I think they will immediately lose 50%+ of individual developer customers and small businesses, large corporations will follow.
Their business looks solid right until it breaks. If OpenAI and Anthropic stop subsidizing heavy usage, people will flee overnight to DeepSeek and whatever good open model is cheapest that week.
I have multiple OpenAI and Anthropic subscriptions and whenever I run out of weekly usage on all of them, I switch to DeepSeek, using it with pi and OpenRouter. The main annoyance here is that the providers on OpenRouter are a mixed bag, some of them I don't trust with my data, some of them have too high prices, especially for cached tokens, others are just not reliable, producing timeouts and other failures. While DeepSeek v4.1 Flash is clearly not on par with GPT-6-Astra or Fable 5.1, it is good enough as a daily driver. The main reason why I still prefer using OpenAI and Anthropic is because they are fast, reliable and cheap enough when using their subsidized subscriptions.
Running inference at scale is hard, and no one is as experienced with this as OpenAI and Anthropic. This is the only real moat they still have. But it's only a matter of time until we will see other providers catching up. At some point LLM inference will be the new shared webhosting.
What I'm buildingDelegate tasks. Get software.
Give Vroni a GitHub issue, bug report, spec, or rough idea. It reads the repo, plans the change, writes code, runs checks, and works toward a review-ready pull request.
Take a look at vroni.com
The comparison with shared webhosting feels right, the model weights are becoming the easy part and running them reliably at scale is the hard part. Your point about OpenRouter providers being a mixed bag on cached-token pricing and timeouts is something I rarely see mentioned, most comparisons only look at the headline price per million tokens. Do you pin specific providers for DeepSeek or let the router pick?