Everyone built the same employee

On Tuesday in San Francisco, OpenAI walked a new agent called Dots onto the DevDay stage. During the demo it failed to respond, per Wccftech. Nobody will remember that. They’ll remember it was the third copy of the same product in seven weeks.

xAI went first. On August 11 it launched Grok Bot, an agent that gets its own computer. On September 8, Meta launched Muse, a personal agent that runs on its own secure virtual machine with its own browser. Ten days later it was the top free app in Apple’s App Store. CNN reported 2.5 million downloads by September 21, per Sensor Tower. Today OpenAI gave each Dot its own cloud computer and browser and connected it to more than 4,000 apps.

Grok, Meta and OpenAI have all shipped it, and we assume Anthropic will do the same. It is the same product, the same user behavior and the same employee.

The product is an agent with a desk, a browser and a login. The behavior is the same everywhere too. You hand it a job, close the laptop and check back later. The employee is the same as well. It keeps working while you’re away, learns how you like things done, and comes back with the finished work.

For three years the labs sold intelligence and let everyone else build the product on top. I was one of the customers, starting in the OpenAI Playground in 2022, where the model finished my sentence. For the past two weeks I’ve used Astra in the latest OpenAI Mac desktop app as my daily coding tool. It is state of the art. That’s my read, not a benchmark. Astra also runs Dots. Now I hand it a task and check back later. That’s the product all three labs just shipped.

The labs have all decided the money is in the whole employee. Answering questions was never going to pay for the data centers. Doing the work is. An agent that finishes the job is something a company can charge a lot for. The direction makes sense.

Now the money. Look at who is missing.

Google owns the inbox, the calendar, the browser, the phone and the documents. It did not ship the agent. Alphabet fell nearly 5% over two sessions as Muse climbed, and DeepMind’s chief said Gemini 4 is coming “much earlier” than planned. The company with the most doors to the house watched someone else walk in.

Microsoft is the second name missing, and OpenAI made that plain almost as an aside. Among the enterprise announcements was a line about specialist Dots, agents built for a defined job inside a company, each with its own identity and credentials. They’re in pilot. OpenAI said it is working with Microsoft to plug them into Agent 365’s governance and security controls. Read that twice. The company that owns Office, Outlook and Teams is providing the badge reader. OpenAI is providing the employee. The rest of the launch points the same way. Space, a shared workspace inside ChatGPT, puts your team, your Dots and Codex on the same files. Dots reach you through Microsoft Teams as easily as through ChatGPT, and per Windows Forum, @ChatGPT in Teams needs no individual licenses. Work starts in ChatGPT. Office is where you export to.

Microsoft did not sit still, though. On September 25 it merged its Copilot apps into one and added Autopilot, a persistent agent that lives in a company’s own cloud tenant. The stock rose 4% to $516 in afternoon trading, and Oppenheimer raised its target to $570. The market’s first read was that things are fine. My read is different. Microsoft has the contract, and the contract is worth a lot. It does not have what OpenAI has, which is 1.2 billion weekly users who already start their work in ChatGPT, plus the model the agent runs on. It may never get that. One keynote is not a verdict. The verdict comes when a procurement team has to pick which agent gets the company badge.

Meta looks fast. Muse launched September 8. Connect was September 23. A Meta enterprise platform followed on September 28. Muse for Small Business landed the next morning. Four moves in three weeks. The limits are real. Amazon has already blocked Muse from its store, and CNN’s reporter found ChatGPT and Gemini sometimes gave more detailed answers. But Meta owns the audience. I ran 45 sites that reached more than 100 million people a month, and I know what an audience does for a product in its first ninety days. It does almost everything.

Google and Microsoft have a harder job than Meta. They have to straddle the enterprise. Their money comes from companies that run on Office and Workspace, and those customers want badges, audit trails and a contract. Every move has to keep that base calm. A company built that way can’t easily ship a bubbly avatar that books your Friday date night, which is exactly what Muse did for CNN’s reporter. The labs sell to the enterprise too. Specialist Dots and Agent 365 prove it.

But the enterprise is not where the interesting part is happening. Something is building underneath it.

I call them the Momentum Generation. They are AI natives who build their businesses and products with AI from day one. Creators. Founders. Inspired people who used to need a team, a budget or a technical cofounder, and now hand the work to an agent. I’ve seen a generation show up before the industry had a name for it. Before they called it the creator economy, I was arming the rebels.

The bottom moved first. Peter Steinberger’s open-source OpenClaw showed the whole shape early this year, and OpenAI hired him in February. OpenClaw now lives in a foundation that OpenAI says it will keep supporting. OpenAI does not own it. Nous Research’s Hermes Agent, open source and running with 300-plus models, has passed 200,000 GitHub stars. The scrappy version came first. The big labs are catching up to what the hackers already built.

Dots, Muse, Grok Bot and Hermes are all an early step in the Momentum Generation’s direction. Early is the right word. CNN’s test of Muse was a reminder that these agents still make mistakes, and Dots failed its own demo. The tools are not finished. They are good enough to start, and the people who start now will be a long way ahead by the time the tools are finished.

That is why I put my chips where I do. Builders can’t afford to be trapped, because the employee they train this year is the asset they carry into the next. Anthropic published MCP and Agent Skills as open standards. Hermes runs on any model and keeps its skills and memory when you switch. OpenAI backs OpenClaw while selling Dots. Dots themselves live in OpenAI’s cloud and run on one model, and I won’t pretend otherwise. Anthropic hasn’t launched its own Dots yet. On September 16 it merged chat and Cowork into one window and added Docs and Slides in beta, and we assume the always-on employee is next.

The enterprise pays the labs, and it will pay for an agent that never fails a demo. The Momentum Generation buys something else, which is leverage, and it will forgive an agent that stumbles on stage. The next companies come from there.

An AI employee is about to join your business. Ask one question first. If you switch to a different one, do you keep what it learned about your business?

Sources:

• OpenAI launches Dots, VentureBeat
• OpenAI DevDay announcements, Runtime Wire
• OpenAI DevDay 2026, Decrypt
• ChatGPT Space and Teams integration, Windows Forum
• Introducing Grok Bot, SpaceXAI
• Introducing Muse, Meta
• Meta says Muse can do things for you, CNN
• Microsoft Copilot overhaul, 24/7 Wall St.
• Anthropic merges chat and Cowork, TechCrunch