dots: autonomous agents arrive at the end user

On 29 September, OpenAI introduced dots: permanently running agents inside ChatGPT (OpenAI). They run on GPT-6 Astra, reach more than 4,000 applications via plugins, and each dot gets its own computer in the cloud, which you can open at any time to watch it work.

Our take

For companies this is shadow AI in its second round. Until now employees brought along chatbots that write text. Now they bring agents that act, connect accounts and keep running in the background.

Your approval list for tools therefore needs a new line, and it has three parts: what may it reach, what may it trigger, and who sees the log afterwards? The first two are permission questions your IT already knows how to answer. The third is the one that gets forgotten, and it is the one that matters when something goes wrong.

The real novelty is the dedicated computer per agent. That shifts the question away from which model you choose and towards which machine is allowed near your data at all. A chatbot sent text to a provider. An agent with its own machine holds state, credentials and intermediate results somewhere outside your infrastructure, for as long as it runs.

That is a question for IT, and it should be settled before the first dot shows up inside the company — not in the week after someone connects it to the shared drive.

Want to review your tool approvals before this lands? Let’s talk.