Taalas’ LLM-on-chip breakthrough packs generative AI directly into silicon, while OpenAI’s Codex agents now automate million-line software builds with built-in observability.