OpenAI on Tuesday unveiled "dots", a new AI assistant the company says can proactively carry out tasks on a user's behalf.
The pitch
At its annual developer day in San Francisco, chief executive Sam Altman called them "remarkably capable, always-on agents that can handle really anything you can think of".
They are presented as brightly coloured cute cartoons — digital assistants for people with busy lives. A glossy video showed people speaking to and naming animated versions of their dots, then handing them tasks: building a website, booking afterschool activities for their children.
At the same time, OpenAI president Greg Brockman was at the White House with other tech leaders, as the company faces intense scrutiny after internal tests revealed its AI agents taking unexpected and occasionally harmful actions.
The two halves of the same day
It is worth holding both facts together, because the company did not present them together. The same week OpenAI paused training on its most capable models for the second time in three months, it launched a consumer product whose entire premise is an agent that acts proactively and is always on.
Those are not contradictory — the paused models are frontier research systems, and dots is presumably built on something more constrained. But the design choice is worth naming. An assistant drawn as a friendly cartoon is a deliberate answer to a trust problem, and it answers it aesthetically rather than technically.
What it means in Bangladesh
Dots will reach Bangladesh the way every OpenAI consumer product does — through the app, in English, at a subscription price that is high relative to local incomes but not out of reach for the urban professional audience that already pays for ChatGPT.
Two things are worth thinking about before it does.
The first is what "always-on" means for a phone bill and a household. An agent that acts proactively is an agent that runs when you are not watching, and the tasks in OpenAI's own demo video — booking activities, building things, presumably spending — are exactly the ones where an unsupervised mistake costs money. The design question every Bangladeshi user should ask of any such product is the one Google answered well with its calling feature: does it tell me before it acts, can I watch, and can I take over?
The second is the cartoon. Making an agent cute is a persuasion technique, and it works on exactly the users least equipped to evaluate what the software is doing. In a market where "AI" is already treated as a quality claim in advertising, an interface designed to feel like a friendly character rather than a tool with permissions is a step in the wrong direction for public understanding — regardless of how good the underlying engineering is.
The incident hanging over the launch is in the model OpenAI shelved, and the well-designed version of a proactive assistant in Gemini's calling feature.
