Same with Codex and very soon with open source & local models. Training great models (for coding and similar tasks) seems to be a question of scale and not much more.

It is likely that 99% of the value created by Anthropic / OpenAI / friends will go the end user. Which is great news.