Technical Context
I didn't latch onto the drama around Anthropic here, but a more down-to-earth thing: in the production cycle, people are simply tired of waiting. And this matters for AI automation, because in real feature assembly, it's not the loudest brand that wins, but the one that doesn't slow my rhythm.
The scenario described is very familiar: I take a feature, run it through analysis, then code, then do the review myself. For this mode, Codex often feels smoother in practice. Especially when you need a fast pass without extra chatter and with good instruction following.
Looking at available comparisons, the picture isn't black and white. Claude Code is strong on large existing codebases, sits deeper in the CLI flow and sometimes proves faster on complex tasks. But Codex regularly wins where predictability, autonomous runs, and fewer surprises in responses matter.
On tokens, there's no single magic truth either. In some tasks, Claude Code is noticeably more economical; in others, Codex spends less. But from the feel in daily work, I understand why some developers are now switching: if the model answers at the right pace and keeps my request context without unnecessary wandering, I forgive a lot.
As for the part about "Anthropic's problems with the government," I'd keep that in the status of community talk, not fact. Same with rumors about Sonnet 5: plenty of discussion, no confirmation. I wouldn't build AI integration or a product roadmap on such leaks.
What This Means for Business and Automation
First: picking a tool for your team shouldn't be based on a benchmark from a single thread. I'd look at your real flow: new features, reviews, cost of errors, token consumption, and cycle speed from idea to merge.
Second: if you have many autonomous tasks and short iterations, Codex currently looks very practical. If you have a heavy legacy context and need tight developer-in-the-loop, Claude Code is still too early to dismiss.
Third: those who lose are the ones waiting for the "perfect model" and not building proper AI architecture around the process. At Nahornyi AI Lab, we solve exactly these junctures for clients: where an agent is needed, where a regular copilot, and where it's not even about changing the model but the workflow itself.
If your team is already stuck on code generation, review, or internal development, there's no need to guess based on rumors. Better to break down your process step by step and build AI solution development under real load. If you'd like, Vadym Nahornyi and I at Nahornyi AI Lab can help you build such a scheme without unnecessary hype and with clear business benefits.