OpenAI's $230 agent keypad: supervision is the product

OpenAI's first physical product is a $230 keypad that sits next to your keyboard and tells you what your AI coding agents are up to. It is easy to dismiss as a merch drop for people with disposable income and a mechanical keyboard habit. But the reason an agent keypad exists at all is a real problem: once you're running several agents at once, you lose track of them, and the bottleneck stops being the model and starts being you.

What actually happened

TechCrunch's hands-on describes the device — called Micro, built with specialty keyboard maker Work Louder — as six frosted "agent" keys across the top, six command keys below, a voice dictation button, and a send button. The agent keys are the interesting part: each one is color-coded to a live status. White means idle, blue means thinking, green means the task is done, red means it errored.

Reviewer Lucas Ropek's verdict was lukewarm — it takes real time to learn, and he questions what it buys you over ordinary peripherals. Coder reaction has been largely dismissive on price, given that a programmable macropad is a well-trodden DIY project. Fair. We'd note that some of the specs floating around in coverage of this device, including a rotary dial for adjusting agent reasoning level, don't appear in the hands-on, so treat those as unconfirmed.

Why this matters for your business

Ignore the hardware. Look at what the hardware concedes.

OpenAI built six status lights because six is roughly how many parallel agents a person can supervise before the whole thing becomes a fog. That's the actual state of agentic development in mid-2026: the models are good enough to run many tasks at once, and the human sitting above them has no dashboard. So a company with a trillion-dollar compute plan shipped a glowing plastic box to solve it.

If you're deploying agents anywhere in your business — coding, support triage, order processing, back-office data entry — the lesson transfers directly and it has nothing to do with buying a keypad. The supervision layer is part of the system, not an afterthought. Every agent you run needs three things visible without you asking: what it's working on, whether it's blocked on you, and whether it failed. If that lives only in a terminal scrollback or a chat thread, you will find out about failures from a customer.

We build this as plain infrastructure, not a product: agents write structured status to one table, a small dashboard reads it, and anything in an error or awaiting-approval state pages someone. It's unglamorous, it costs a fraction of the agents themselves, and it's the difference between agents that reduce work and agents that generate a new kind of work — the kind where you're constantly checking on them. A $230 keypad is one company's answer. Yours should be a screen your whole team can see.

Sources: TechCrunch.