Beyond the Buzz: How OpenAI’s ‘Dots’ Agent is Transforming AI from Assistant to Advocate

Hello again, and welcome back to Fast Company’s Plugged In.

In recent years, the rapid evolution of artificial intelligence has moved at a pace that is, frankly, difficult to keep up with. Becoming "gobsmacked" by the ever-expanding capabilities of large language models has become my new normal. It is a recurring state of surprise that has punctuated the last few years of technological development. But perhaps more importantly, I have found this technology becoming increasingly essential to my own productivity. From drafting emails to refining code and synthesizing complex research, AI tools have shifted from being novel experiments to being integrated components of my daily workflow.

Yet, for all the utility these models have provided, they have historically functioned as reactive tools. They wait for a prompt, execute a task, and then wait for the next instruction. What AI never did—until just last week—was save my bacon.

That dynamic shifted significantly when I provided OpenAI’s new agent, known as Dots, with access to my calendar and my email communications. The difference was immediate and, in a practical sense, transformative. Shortly after granting the agent these permissions, it alerted me to a discrepancy it had identified between my scheduled commitments and my correspondence. I had manually entered a speaking engagement on one date in my digital calendar, but a confirmation email buried in my inbox mentioned a completely different day.

As it turned out, the event organizers had rescheduled the engagement and had failed to notify me directly. If I had relied on my calendar alone and shown up on the original day, I would have walked into an empty room, wasting time and risking a professional embarrassment. Dots caught the error by cross-referencing my static schedule with the fluid reality of my email flow. It didn’t wait for me to ask, "Check my schedule for errors." It acted as a proactive monitor, sensing that the two sources of truth did not align.

This experience represents a shift in the philosophy of AI adoption. For much of the last two years, the industry has been obsessed with "agentic AI"—a term that has often felt like a vacuous buzzword, thrown around at tech conferences to signal innovation without necessarily delivering a tangible change in user experience. However, after having Dots proactively bring several significant, unprompted facts to my attention, the technology is starting to feel less like a marketing trend and more like a core, indispensable benefit of weaving AI into a professional workflow.

The official unveiling of Dots occurred on September 29 during OpenAI’s annual DevDay conference, hosted in San Francisco. The event, which followed a period of intense speculation and the company’s decision to shelve certain model releases over safety concerns, served as a platform for OpenAI to demonstrate its next phase of product development. The message from the stage was clear: the company is moving beyond simple text-in, text-out interfaces toward autonomous agents capable of navigating the complexities of human digital life.

Currently, the availability of Dots is restricted. It is locked behind the paywall of ChatGPT’s most premium tiers, with access starting at $100 per month. This high barrier to entry serves a dual purpose for OpenAI: it allows the company to test the technology with a smaller, highly engaged user base while it continues to refine the agent’s capabilities and, crucially, to determine the long-term economics of operating such a resource-intensive service.

The technical infrastructure powering Dots is substantial. It operates on OpenAI’s highly capable GPT-6 Astra model, which was announced earlier in September and represents the current pinnacle of the company’s research. What makes the architecture of Dots particularly unique is that it doesn’t just run a script; it essentially provides each user with a dedicated Linux computer in the cloud. This environment allows the agent to execute code, manage files, and interact with various applications in a way that mimics a human user sitting at a workstation. By granting the agent a "computer" of its own, OpenAI is providing it with the sandbox necessary to solve problems that require more than just linguistic reasoning—they require the ability to manipulate digital environments.

The implications of this transition toward agentic behavior are profound for the future of productivity software. We have spent decades learning how to use software—learning the specific interface of a calendar app, the file structure of a cloud drive, and the search syntax of an email client. Agents like Dots promise to invert that relationship. Instead of the human learning the software, the AI learns the user’s habits, preferences, and sources of information. When the AI can monitor these streams of data, it ceases to be a tool that needs to be "opened" and becomes an ambient partner that resides in the background, ready to interject only when a conflict or opportunity arises.

Of course, this level of access brings with it a new set of considerations. Granting an AI access to one’s private calendar and email is a significant step that requires a high degree of trust in the platform’s security and privacy protocols. As OpenAI looks to take agentic AI "wide"—moving from this exclusive, high-priced testing phase to a broader public rollout—the company will have to navigate not just the economic challenges of compute costs, but also the societal questions surrounding digital autonomy and data privacy.

The company’s roadmap suggests that once it has adequately refined the performance of Dots and established a sustainable business model, we can expect to see these capabilities integrated into more accessible versions of their products. For now, the "Dots" experience remains a glimpse into a future where the friction of digital administration—scheduling, cross-referencing, and coordination—is handled by a system that understands the context of our work as well as we do.

For those of us who have spent the last two years testing, critiquing, and integrating AI into our professional lives, the emergence of agents that can actually save us from ourselves is the most significant development yet. It turns the AI from a sophisticated parlor trick into a genuine assistant. While the $100 price tag and the current constraints of the beta phase keep it out of reach for the average user for now, the path forward is clear. The era of the "agentic" web is no longer a hypothetical scenario discussed at industry summits; it is a live, functioning reality that is already busy managing the details we are too distracted to notice.

As I look at my calendar now, I find myself checking it with a different perspective. I am no longer just looking for the next appointment; I am looking at a ledger that is being audited in real-time by an intelligent system. It is a strange, yet welcome, sensation to have a piece of software looking out for me, rather than just waiting for my next request. If this is the trajectory of where AI is headed, the promise of "increased productivity" may finally move beyond the abstract and into the daily, tangible reality of our working lives. Whether or not it will remain a luxury or become a standard utility remains to be seen, but the utility itself is now undeniable.