@theo Enjoy it while it lasts. The cycle: lab ships the best model → everyone switches → demand outruns compute → limits tighten or quality slips → users bail → the lab they left has spare GPUs to train the next best model. Repeat.
@thsottiaux Dot has a routing problem. Every request goes to the same Codex thread on Astra, whether it needs it or not. It burns through a 20x plan like nothing. Real agent efficiency means matching the task to the model.
@martyamark Have your agent write blog posts so other agents can write blog posts citing your blog, which your agent can then cite in another blog post.