Personal AI agents have existed as a concept for years. What’s changing now is that they’re starting to work, in ways that are specific, useful, and occasionally surprising. The arrival of OpenAI’s Dots agent, alongside competing products from Meta and a startup called Instinct, marks a shift worth paying attention to: agentic AI is moving from demonstration to daily use.
From Prompt to Proactive: What Agents Actually Do Differently
Most AI tools are reactive. A user types a question, the system answers. Agents operate on a different logic. They are given access to a person’s accounts and data, and they act on that information without waiting to be asked.
Dots, announced by OpenAI on September 29 at its DevDay conference in San Francisco, illustrates this clearly. When connected to a user’s calendar and email, it does not simply retrieve information on request. It monitors, cross-references, and flags discrepancies. In one documented case, Dots identified a mismatch between a calendar entry and a confirmation email, catching a rescheduled event that the organizer had failed to communicate. That kind of catch requires reading across multiple data sources simultaneously and understanding what the inconsistency means for the user.
This is what most coverage of AI assistants misses. The capability gap is not about generating text faster. It is about shifting from a tool that answers to one that notices.
A Crowded Field, With Real Differences
Dots is not alone. Meta’s Muse agent is currently among the most downloaded iPhone apps. Instinct, a startup product that operates inside messaging platforms like WhatsApp and Apple Messages, carries a reported valuation of $10 billion and still has a waitlist. Hark Pro, another entrant, launched on October 6 and is available for free.
All three share a basic architecture: connect to user accounts, learn from the data there, and surface relevant information or actions. The differences emerge in execution. Dots, which runs on OpenAI’s GPT-6 Astra model and provides each user with a cloud-based Linux environment, demonstrated capabilities that the others did not replicate in comparable testing. It pulled spending data from emailed receipts to build a wireless plan comparison without being asked for those details. It identified a neglected message from a freelance author buried in an inbox. It found a discount on a product the user was already considering.
Muse and Instinct did not perform the same level of inference from scattered data. That gap matters because it points to what separates a useful agent from an impressive demo: the ability to synthesize incomplete information into something actionable.
Dots is currently available only to users on ChatGPT’s higher-tier plans, starting at $100 per month. OpenAI has stated plans to expand access more broadly once the product is further refined. CEO Sam Altman has described a future where agentic AI reaches billions of users.
What This Means Beyond the Product Announcements
The deeper significance here is structural. Bill Gates wrote in 2023 that agents would eventually dissolve the category of discrete apps by consolidating functionality into a single conversational layer built around deep knowledge of the individual user. That prediction is beginning to look less speculative.
When an agent can handle tasks that previously required specialized software, such as processing PDFs, researching purchases, triaging email, or cross-checking schedules, the question of which apps a person needs starts to shift. This does not mean software disappears. It means the interface layer changes, and with it, the relationship between users and the tools they rely on.
For individuals, the practical implication is access to a kind of ambient attention that was previously unavailable. An agent that monitors your accounts and surfaces what matters, without being prompted, handles a cognitive load that most people carry silently. The value is not in any single action but in the accumulation of small catches and nudges over time.
For organizations, the implications are broader. Agents require access to data to function. The more accounts and systems they can connect to, the more useful they become. That creates real decisions about what access to grant, to which systems, and under what conditions. The user experience questions are also unresolved: current implementations involve navigating between multiple sub-versions of the same app, and the boundaries between them are not always clear.
OpenAI’s product chief Thibault Sottiaux has indicated that the company views Dots as a step in an ongoing process rather than a finished product, and that the distinction between Dots and ChatGPT may narrow over time.
In Short
Agentic AI is not a future category. It is a present one, with real products, real limitations, and real consequences for how people manage information and attention. The shift from reactive to proactive AI is the meaningful change here. Whether any current product has fully solved the user experience is a separate question. What is clear is that the underlying capability, an AI that notices things you didn’t ask it to look for, is no longer theoretical.
Based on reporting from Fast Company - Tech.