OpenAI's Codex Push Feels Familiar, and That's Not Necessarily Good

Every few years, something comes along that's going to change everything. PLCs were going to eliminate electricians (they didn't). MES systems were going to give us perfect visibility (sort of, eventually, after years of painful integration). Digital twins were going to let us simulate everything before building it (still waiting on that one to fully deliver).

The technology usually works in demos. It works in controlled environments with clean data and motivated pilot teams. Then it hits the real world, where legacy systems don't talk to each other, where the guy who understood the old setup retired last year, where there's always some weird edge case nobody anticipated.

I called my old colleague at Siemens last week, actually. He's been watching the AI coding tools closely because they're dealing with the same questions we had about automation: what happens to the people? He said something that stuck with me. "The tool works fine. It's the organizational change that kills these projects."

What Actually Matters

The Codex SDK and admin tools are, in a way, more interesting to me than the AI itself. Usage dashboards, workspace management, enterprise controls. That's the boring stuff that determines whether something actually gets adopted or becomes shelfware.

When we rolled out new robot programming environments at Kuka, the flashy features got the press releases. But adoption came down to: could the maintenance techs learn it? Did it integrate with the existing PLCs? Could we train people without shutting down production for a week?

OpenAI seems to understand this, which is why they're partnering with the Accentures and PwCs of the world. Those firms know how to do enterprise change management. They've been doing it for decades, sometimes well, sometimes not.

The 4 million weekly active users number is interesting but I'd want to know more. How many are actually shipping code versus just playing around? What's the retention look like after the novelty wears off? OpenAI didn't disclose those figures, at least not in what I read.

My Actual Take

I'm not bearish on this stuff. AI coding assistants probably will change how software gets made, same way CAD changed mechanical design and simulation changed how we validate robot paths before cutting metal.

But the breathless announcements remind me too much of every other hype cycle I've lived through. The technology is usually real. The timeline is usually wrong. The organizational challenges are usually underestimated.

If you're an enterprise thinking about rolling this out, my unsolicited advice: start small, pick a team that's motivated, measure what actually matters (code shipped, bugs found, time saved), and don't believe the vendor's projections. That advice hasn't changed since I started in this industry, and I don't expect it to change now.

The Figma integration is neat, though. I'll give them that. Connecting design and code has been a mess forever. Whether this actually solves it remains unclear, but at least someone's trying.

Aisha Patel · 44 mins ago · 8 min