Here's the thing everyone got wrong this week: they saw a $230 keyboard and assumed OpenAI was suddenly serious about hardware.
It isn't. Not with this, anyway.
The Codex Micro is a novelty. OpenAI told TechCrunch it's a "limited-run collaboration," which is corporate for "don't build your workflow around this." Work Louder's own store page hedges with "limited quantity." It's already sold out on OpenAI's site.
The actual hardware story is the one hiding two clicks away — a screenless, portable smart speaker with "mechanical elements that can move on their own," reportedly designed by ex-Apple engineers, which per TechCrunch is part of why Apple is now suing OpenAI for trade theft. That device is the long game. The keyboard is a desk ornament with good lighting.
The actual hardware story is a screenless smart speaker with mechanical elements that move on their own. That device is the long game. The keyboard is a desk ornament with good lighting.
So why lead the week with a keyboard? Because it tells you something the lawsuit doesn't: what OpenAI thinks its product actually is now.
Last week we asked whether ChatGPT Work survives as its own thing. This week OpenAI answered by collapsing the whole question. ChatGPT, Codex, and ChatGPT Work are now one app — the "superapp" the company has been promising — and the Codex Micro is a physical control surface bolted onto it.
Sit with that. OpenAI's first branded piece of hardware isn't for chatting. It's for managing fleets of coding agents.
Here's what the thing does. Six frosted keys light up to show agent status, reportedly including states like thinking, running, waiting, and finished — so you can glance at your desk instead of clicking through conversations to find out what your bots are up to. Programmable keys handle frequent moves like accepting a change or branching a thread. A joystick flicks between workflows like reviewing a pull request or debugging.
And then there's the dial.
The rotary dial adjusts how hard Codex thinks. Turn it down for fast answers on simple tasks; crank it up when something needs heavier reasoning and more compute. It turns AI reasoning — normally buried in a settings menu — into something closer to a volume knob.
That's genuinely clever, and it's the most honest object OpenAI has shipped. It admits, physically, that "thinking harder" costs time and money, and puts that tradeoff under your thumb.
But should you buy it? Almost certainly not, and OpenAI basically agrees.
Thirteen mechanical switches, a joystick, a touch sensor, and a dial is a lot of hardware to dedicate to one AI coding tool. At $230, its value depends entirely on how deep Codex sits in your daily routine. If you're orchestrating several agents at once, permanent buttons and status lights might genuinely earn their keep. If you use Codex occasionally, this is $230 of very specific convenience you'll admire and never justify.
The tell is in the framing. Engadget notes the whole superapp "can be used without dedicated controls" — the keyboard is for "diehard users." This is an accessory for the top 1% of power users, sold to make everyone else feel the platform is deeper than it is.
Which brings us to the more consequential news, the stuff that ships to everyone.
First, search. ChatGPT's sidebar search used to surface only past conversations, leaving your uploaded files, projects, and generated images stranded. Now one search covers all of it, with filters by content type — pull only images, or only files tied to one project — across web, iOS, and Android, on every plan including free.
It doesn't change what ChatGPT stores. It just makes what's already there findable in seconds. That's a small feature with a big tell: OpenAI now assumes you're living inside the app long enough to lose things in it. You don't build account-wide search for a chatbot. You build it for an operating system.
That's the real superapp thesis, and OpenAI's own Codex and ChatGPT app lead, Andrew Ambrosino, said it plainly to Big Technology: "Anything that you can do on a computer, you can now do with this app." People are editing videos, building spreadsheets and slide decks, writing production code.
There's still a toggle between Codex and ChatGPT Work, but Ambrosino insists it changes only the experience, not the capability — "The toggle does not affect what this product can do." When asked why the app doesn't just infer your intent, he was candid: "There are a lot of use cases where we can't necessarily tell what you want."
He was equally candid about arriving late. Anthropic got here first with Claude Code and Cowork bundled into one app. Ambrosino's spin: "Going first has its weaknesses sometimes. We have seen how this stuff should work, and now we can go execute on it."
Whether they've executed is contested. One reviewer in Big Technology's own link roundup concluded the superapp "doesn't yet hit the mark." Consolidation is not the same as coherence.
Then there's the third thread this week, and it's the one that matters most beyond developer desks: teens.
OpenAI rolled out real safeguards, and its argument is worth quoting because it reframes the whole debate. The company says nearly 90% of teens already use ChatGPT weekly, and that locking them out entirely would leave them "unprepared for one of the defining technologies of their generation."
So instead of a wall, they built guardrails. ChatGPT now uses age prediction to guess whether a user is under 18, and — critically — when it's unsure, it defaults to the safer teen experience. That's a meaningful design choice: err toward protection.
The teen mode adds stronger safeguards against graphic violence, self-harm, unhealthy body image, and risky viral challenges. Teens get more frequent break reminders on long sessions.
Parents get Quiet Hours, the ability to disable Voice Mode, control over image generation, and notifications in high-risk situations — including signs of potential self-harm. There's also Study Mode, built with teachers, that walks students through problems step by step instead of handing over answers, and parents can now set it on by default.
This isn't out of nowhere. It's OpenAI executing the Teen Safety Blueprint it published back in November 2025, complete with the line, "We aren't waiting for regulation to catch up." The company would rather set the standard than be handed one.
Read the age-prediction system charitably and it's responsible design. Read it skeptically and it's, arguably, a company that now has enough insight into your behavior to guess your age — the same capability that makes account-wide search possible, pointed in a different direction.
So step back and look at all three threads together, because they're the same move.
The keyboard says: we have power users deep enough to buy dedicated hardware. Search says: you live in here now, so we'll help you navigate. Teen safety says: even minors basically live in here, so we'll manage that too.
Every announcement this week assumes ChatGPT is no longer a thing you visit. It's a place you inhabit — for coding, for schoolwork, for running your computer.
The keyboard is the toy. The superapp is the product. And the teen controls are the quiet admission of how thoroughly this thing has already embedded itself in daily life.
Don't buy the keyboard. Do pay attention to what it's confirming.