Dots, Grok Bot and the AI agent race: who is going to win it

On September 29, at its DevDay in San Francisco, OpenAI introduced Dots: agents that work all the time, without waiting for you to write to them. Each Dot runs on GPT-6 Astra, has its own cloud computer with a browser, learns from your corrections and connects to thousands of apps. You talk to it from ChatGPT, Slack or Teams, and OpenAI already said specialist Dots and teams of Dots working together are coming.

It did not arrive alone. Weeks earlier, Meta had launched its Muse series of agents. And on the same night as DevDay, xAI bought the dot.com domain and pointed it straight at the page for Grok Bot, its own agent. Nobody read that as a coincidence.

Grok Bot goes in the same direction: named bots with memory and their own computer that keep working when you log off. On September 28 Elon Musk announced they now work as a team, handing tasks and context to each other, and two days earlier xAI connected them to bank, card and investment accounts.

What changed is not that AI answers better. It stopped answering and started doing. A chatbot tells you how to reply to an email; an agent replies, books the call and tells you when it is done.

For a small business that is what matters, because its problem was never a lack of answers. It is a lack of hands: the follow up nobody does, the messages left unanswered over the weekend, the blog that has not been updated since March.

That is why every big player wants its agent. Whoever owns the place where people delegate their daily work owns the most valuable relationship of the next decade.

Dots and Grok Bot are general agents: they can do almost anything, which also means they arrive blank. You define the job, the rules, the tools and what a good result looks like. For a technical team that is power. For the owner of a clinic or an online store, it is one more project.

Marblism did not arrive at agents this week. It was talking about AI employees when the rest of the market was still talking about chatbots, and that is still its only product: an AI team with clear roles for small businesses.

Today there are eight roles: Penny writes the SEO, Stan finds clients, Rachel is the receptionist, Sonny runs social media, Eva is the executive assistant, Alisson closes, Walter builds websites and Linda is the legal assistant. More than 40,000 businesses use them, they are rated 4.8 out of 5 on Trustpilot and the company is backed by Y Combinator.

The difference is not the model, which anyone can use. It is that each employee already knows its job, and that behind them there is a company touring cities with events, live sessions and a partner network so business owners walk away with their employees up and running.

Here is the real question. For OpenAI and for X, agents are one more line on a huge list: models, video, devices, social networks, cars, finance. They have the muscle and the best models, and that is good for everyone, because companies like Marblism build on top of those models.

But when a product is one of twenty, its priorities change with the quarter. When it is the whole company, every decision is made with the customer who depends on it in mind.

The race will not be won by whoever announces the loudest. It will be won by whoever works for its community: the company that understands a small business, helps it implement and is still there when something breaks. For some, agents are the business. For others, they are one more project.

Whatever the option, these are the questions we ask before putting an agent or an AI employee into a business:

To be clear: at KOS we are Marblism partners and we implement their AI employees in clinics, real estate agencies, law firms and stores. Keep that in mind as you read this. We also use Claude, ChatGPT and Grok every day, and we will test Dots as soon as it reaches our accounts. The race is just starting, and our clients benefit from everyone running it.