We're so focused on AGI that we're missing so many opportunities of building task-specific models that are incredibly effective in practice. I've asked for something like this a long time ago, and I'm glad someone finally shipped it. Perhaps it is time to write my own Codex where: - I write the next task - a grep model gathers the relevant context - the smartest coding model writes the code - repeat At least intuitively I think that will be more effective than the "CLI coding chat" interface we have right now. Think about it, what chatting really gets us? Some persistency across different tasks? It stops being good after 2-3 iterations so we have to /compact anyway.