OpenAI Dots: Proactive Assistant, Controls, and Payments

OpenAI is launching dots, a personal agent connected to ChatGPT designed to act without solicitation. The company states it is in discussions with payment processors and promises finer privacy controls, as the application sometimes requires extensive access. Offered to Pro subscribers starting at $100 per month, the assistant aims to anticipate needs, but its adoption and interaction with third-party actors remain open questions.

OpenAI discusses with payment processors and promises finer controls

Alexander Embiricos claims that OpenAI is engaged in numerous discussions with payment processors, while acknowledging that the company is not always the fastest to deploy new capabilities and does so in collaboration with the ecosystem. He believes it is necessary to build very granular privacy controls and reports that the launch has sparked criticism related to conservative default settings and frequent permission requests, with the intention of adjusting them. OpenAI indicates that dots retains the controls of ChatGPT, including the ability to disable model training and manage the resources the agent accesses, but feedback points to a lack of granularity and sometimes insistent behavior, such as refusing read-only access to cloud drives. Embiricos also highlights the intelligence and discretion of the model. Payment partnerships could allow the assistant to make purchases, a prospect presented as potentially lucrative and a significant demand, while access to email is already demanding and access to bank accounts remains excluded for now.

A proactive agent that anticipates, with timing still improvable

Dots is available in the desktop and mobile applications of ChatGPT for Pro subscribers or higher, with offers starting at $100 per month. Connected to calendars, email, computers, and other services, the agent attempts to assist in completing tasks, notably through a proactive search function that regularly reassesses what it knows and the accesses it has to propose useful actions. Designed as a persistent agent, it operates without solicitation, which, for OpenAI, translates to increased token usage. Alexander Embiricos describes it as proactive, although tests report sometimes poorly timed anticipation. He mentions a conflict warning that arrived three minutes before an event early in development and expects improvements in timing. The proactive approach is deemed central, illustrated by a case where a dot detected and analyzed a failure before an engineer and provided useful information.

From coding tools to a unique companion in ChatGPT

Alexander Embiricos explains that a significant portion of dots' functions was achievable with the Codex application, launched earlier this year, but at the cost of more work and support. According to him, OpenAI has evolved ChatGPT Work from agentic coding tools developed for engineers and designed dots as a unique entity with ongoing conversation for greater intuitiveness. The name plays on the negative space of the OpenAI logo with the ambition of a potentially omnipresent presence; Embiricos says he learned late that Elon Musk owns the dot.com domain. Dots was unveiled during the developer day at the end of September. Embiricos, who leads the enterprise product team, detailed these choices at TechCrunch Disrupt, discussing a "convergent evolution" of efforts around assistants and the importance of user presentation.

Uncertain adoption and data debate in the face of Muse and alternatives

The rise of AI agents faces questions about their adoption: some companies are not ready to engage with agents, and user aversion to these interactions has been reported. Alexander Embiricos says he shares this reluctance, having a human assistant because OpenAI partners want a person in front of them and not considering meetings with others' agents. While players like Instinct are presented as having experienced rapid growth and Meta promotes Muse, the debate also revolves around the data necessary for the usefulness of these tools. Muse has been criticized for building personal profiles that could also serve Meta's advertising business. Meanwhile, the startup Hark emphasizes that it does not collect data for advertising purposes, establishing a contrast with the aforementioned giants.