OpenAI DevDay: the 'Dots' personal agents and the (partial) end of the laptop

By Turing Post · September 29, 2026.

The editorial in this issue of FOD opens with a personal anecdote from the author: during a night-time Uber ride through San Francisco, in a Tesla with autonomous driving engaged, the driver asked her why she was still using a laptop while the car drove itself. His answer was revealing: he barely uses a computer any more, because all his agents live on his phone. The author connects this scene with the recent craze for buying Mac minis to host 'OpenClaws' that need to be always connected, and suggests that era may be on its way out.

The reason for raising the question is OpenAI's DevDay, held that same day at Fort Mason (San Francisco), where the company unveiled 'Dots', its new generation of personal agents. As explained in the email, Dots have their own computers in the cloud, their own browsers and connections to the user's applications. OpenAI claims these agents can keep working on projects continuously and maintain context across ChatGPT, Slack and Teams, with support for text messages announced as the next step.

The author admits to being a regular user of OpenAI's products, but confesses that she is keeping her 'OpenClaw' because, in its current configuration, it is practically free for her and runs on a physical computer she can switch off; she values having that option. Even so, she considers Dots a very interesting development for OpenAI, especially because the company has had to postpone the launch of its next model, Astra 6.1. She notes, with a certain irony, that perhaps that model is not even needed yet, given how broad the available capabilities already are, and that the desirable thing would be for those capabilities to go towards delegating routine tasks to an always-available assistant with 'character', which she compares to a Clippy arriving in the right century.

One point the author looked into directly was routing between models for Dots, since users may find it confusing. As she tells it, Tibo —whom she refers to as 'The Reset Guy'— replied that, for now, Dots runs on the Astra model because they do not want to compromise service quality. Tibo added, verbatim: "We're working on routing. We didn't announce routing today, but this is something that we're working on." He also commented that, looking ahead, OpenAI believes simplicity will end up prevailing over complexity.

One detail noted in the email is that conversations with Dots do not consume tokens, something the author views positively. She also reflects on the amount of infrastructure OpenAI maintains behind this product: an agent needs files, tools, permissions and a place to run, as well as enough preserved context that picking a project back up does not mean explaining everything again from scratch.

The email lists the rest of the DevDay announcements in that same context: 'Cloud Codex', which provides development environments; shared spaces and documents that let people and agents work together; 'Events', which allows events from connected applications to trigger tasks automatically; and cheaper models, such as the newly announced 'Sol 6.1', which make sustaining this kind of continuous work less expensive. In addition, OpenAI would allow subscription fees to carry over to participating third-party tools, would offer managed agents through AWS, and would work with Microsoft on enterprise governance. The author's reading is that OpenAI is seeking a significant role in users' work even when they are using other companies' software.

The text places OpenAI within a broader industry trend: it mentions that Anthropic is pursuing related ground with Claude Code, Cowork and managed agents; that Google and Microsoft already hold much of the information from users' work environments; that Meta's Muse approaches personal assistance through commerce, communication and glasses; and that Grok is said to be well received by many users. According to the author, all these companies share the challenge of turning access to information into useful actions without requiring constant supervision.

When she asked an OpenAI employee how Dots differ from a product like Muse, he replied that they are similar but that Dots are far more proactive: an assistant capable of comparing and analysing future needs, including tasks the user had not even thought of assigning. The author reflects on the freedom this could bring, comparing it to her own habit of making to-do lists so as not to lose sight of their connections, and suggests that delegating that work to an agent is appealing.

On access, the email details that, for now, there are limits: Dots has not been announced for users on the Free or Plus plans. The Pro plan includes an allowance for deeper work, while the higher tiers offer greater capacity and speed. Getting started would require less technical knowledge than before, but how much work can be delegated depends increasingly on the subscription purchased. It is mentioned that Sam Altman promised to extend access to the mass market soon, and the author says she understands that the company wants more control over such a powerful feature first.

The editorial closes by returning to the opening anecdote: the Uber driver had already found an arrangement that worked for him with his agents on his phone, and the author says she wants to find out whether products like Dots can offer the same to people with no intention of setting up an agent themselves, and whether, after paying for that convenience, they really do end up spending less time managing them.

Outside the editorial, the email also includes a section sponsored by MongoDB on unified versus split database architectures for AI applications, and a 'This Week in Turing Post' block with the schedule of the week's own content (a guide to 'World Action Models', a piece on Gaussian Splatting and an instalment of its 'Library' newsletter on synthetic data platforms), plus a call to subscribe to the Premium version.