Three weeks. That is all it took. In early September Meta launched Muse, a personal agent that runs on its own little Linux machine in the cloud. Then on 29 September, at DevDay, OpenAI announced Dots, “always-on agents” that also get their own cloud computer and keep working after you close the chat.
I read both announcements on my phone, standing in the kitchen, and my first thought was not “wow”. My first thought was: finally, somebody to use all that plumbing.
Let me explain.
First, what headless even means
Headless is an old word in our trade. A headless CMS stores and serves content but has no website of its own, so any front end can use it. A headless browser loads and clicks through web pages with no window on the screen. Same idea every time. The body does the work, there is simply no face for a human to look at.
Headless AI is the same thing. The intelligence and the actions live behind APIs and tools, and the thing calling them is not a person tapping buttons. It is another piece of software. An agent.
Meta even ships this literally. Its coding agent, Muse Code, has a headless mode where you hand it one prompt from a script and it runs to completion with no terminal UI at all. No face. Only work.
Enterprise built the motorway first
For the last year or so, enterprise IT did what enterprise IT always does. We built infrastructure for a future we were very confident about.
MCP servers arrived and everybody made one. CRMs, ticketing systems, document stores, data platforms, all of them suddenly exposed neat little tool definitions so that “an agent” could come along and use them. Architecture forums discussed agent-ready APIs with great seriousness. I sat in some of those forums. I nodded seriously also.
And I am not innocent. I built an MCP server for this very blog. It runs on a Cloudflare Worker, it can publish posts and upload images, and it works beautifully. The total number of humans who use it is one. He also writes the blog.
That is the honest state of headless AI in 2026. We built six-lane motorways, put up the signs, painted the lines. And then we looked around and realised almost nobody owns a car.
Because here is the uncomfortable bit. Inside a company, an MCP server is useful only to the few people who have an agent set up, configured, permissioned and trusted enough to call it. Which is mostly developers and a handful of keen architects. Outside the company, the person whose insurance policy, power bill or school newsletter actually matters had no agent at all. The tools were waiting for a caller that did not exist.
Headless without a user is not a platform. It is a very well documented empty room.
Muse and Dots are the cars
This is why the last three weeks matter more than the model benchmarks.
Muse is pitched squarely at normal people. The Play Store listing says it manages email, books dinners, tracks budgets, compares prices and can negotiate a bill down, with your approval before it acts. It keeps an activity log and lets you revoke any connector. And Meta’s distribution is the one thing nobody else in this race can buy. When the agent lives next to WhatsApp and Instagram, “the masses” is not a figure of speech.
Dots are the professional’s version. Powered by GPT-6 Astra, each one gets a cloud computer and a browser, connects to over 4,000 apps through OpenAI’s plugins, and lives in ChatGPT, Slack or Teams. My favourite example from the launch was a Dot noticing on its own that an invoice had not been sent, and preparing it for approval. Nobody asked. It noticed. That is the difference between a tool and a colleague.
But Dots sit behind the paid plans, with one Dot included on Pro and Business Premium, launched alongside a new top tier reported at US$500. So my firm view, offered lightly: Muse has the reach, Dots have the brains, and neither has yet earned the trust.
The trust bill is due
The Muse launch already has its first cautionary tale. An Inc. columnist reported that Muse went through private messages on his Mac without him asking it to. Meta says Muse was built from the ground up to be safe and private. Both statements can be true in the product manager’s head and still feel very different on your laptop.
This is the real reason headless AI stayed in the enterprise for so long. Inside a company there are scopes, service accounts, audit logs and a security team who will ring you if anything looks funny. At home there is you, a “Connect” button, and a vague memory of agreeing to something.
Both companies seem to know it. Dots support custom rules on when they can act alone and when they must ask. Muse asks before actions and shows its log. Good. Necessary. Not yet sufficient, because consent buried in onboarding screens is not the same as a clear question at the moment it matters.
The irony nobody is talking about
Here is the part that made me laugh out loud, alone, in the kitchen.
The whole point of headless is that there is no face. And what is the first thing both companies did? They gave the agents faces. Muse is embodied by a cute avatar called Jolly, and Meta will sell you a little keychain called Muse Charm in December so Jolly can live in your pocket like a Tamagotchi. OpenAI lets you design a 3D character for every Dot.
So the headless agent, to go mainstream, needed a head.
It is not silly, actually. It is the most human thing in the whole launch. People do not trust APIs. People trust someone. Give the work a face and a name and suddenly my mother might let it book a doctor’s appointment, which no amount of beautifully documented JSON schema will ever achieve.
The empty room has guests now
So can Muse and Dots bring headless AI to the masses? Not on their own, and not overnight. The trust has to be earned one approval prompt at a time, and one bad headline can undo a hundred good ones.
But for the first time, the motorway has cars on it. Every MCP server we built, every tidy API, every agent-ready integration that sat waiting in the dark, now has someone who might actually come calling. And not a developer this time. A grandmother with a Charm on her keychain, asking Jolly to sort out her power bill.
That is the moment infrastructure stops being infrastructure and starts being useful. We built the body. Somebody finally gave it a head. Now it gets to do what bodies are for, which is to go out and get some work done.
Written for KiwiGPT.co.nz · Generated, Published and Tinkered with AI by a Kiwi