ChatGPT Mobile Launches DOT: How Your Pocket AI Agent Gets Work Done For You

OpenAI's ChatGPT app is bringing "proactive AI agents" to your pocket. DOT — an always-on, proactive agent that handles tasks on your behalf — was previously limited to desktop and web, but has now landed on both iOS and Android. Alex from OpenAI's mobile team showcased the setup process and real-world use cases in an official demo.

What Is DOT?

DOT is defined as an "always on, proactive agent that just takes off your plate" — an intelligent agent that stays active and proactively handles tasks for you. Unlike traditional chatbots that passively wait for instructions, DOT is built around the idea of proactivity: it doesn't just answer questions, it uses the context you grant it to execute tasks and drive progress forward.

The value of this kind of agent is heavily tied to the context it has access to. Alex repeatedly emphasized in the demo: the more information you connect to DOT (like your calendar and email), the more it can take off your hands. In other words, DOT's capabilities are largely bounded by how much access you're willing to give it.

Agentic AI is one of the core directions in the evolution of large model applications today, and it's fundamentally different from traditional question-and-answer AI. Traditional chatbots operate in single-turn or multi-turn conversation mode: user inputs → model returns text → interaction ends. Agentic AI introduces a "perceive–plan–execute" loop: the model doesn't just generate text — it can call external tools (such as code executors, search engines, and API interfaces), break tasks down into sub-steps, execute them sequentially, and dynamically adjust its strategy based on intermediate results. The technical foundation underpinning DOT is OpenAI's Codex cloud-based code execution environment and an OAuth-based authorization system for third-party app integrations. That authorization means DOT can read and write calendar events, send and receive emails, and post Slack messages on the user's behalf — operations that are technically nearly equivalent to the user doing them manually, which is why permission management matters just as much as the features themselves.

Mobile Setup: Nearly Zero Friction

Enabling DOT on mobile has been deliberately streamlined to the point where "it almost sets itself up." The steps are minimal:

• Open the ChatGPT app
• Tap the "Meet DOT" entry point in the top-left corner
• The system automatically creates a conversation thread with DOT
• Follow the prompts to grant permissions for your calendar, email, and more

Once setup is complete, DOT gets to work. Alex recommends connecting both your calendar and email, again emphasizing that more context equals better performance.

Personalization: Giving Your AI an Identity

DOT supports a meaningful degree of personalization. Users can give it a name and set a custom avatar. In the demo, Alex named his DOT "Flossie" and created a custom blue avatar (you can also choose an existing CodexPet character). This anthropomorphic design is clearly intended to reduce the distance between users and their AI agent — making the "personal assistant" feel more like a collaborator with its own identity.

Live Demo: How DOT Handles Complex Tasks

The most compelling part of the demo is DOT's actual execution capability. Alex mentioned that DOT can access not only "its own computer" (a cloud compute environment) but also the user's computer, enabling it to tackle complex work like coding and prototyping.

Alex then gave Flossie a multi-step task via voice: prototype an iOS home screen widget that displays DOT's avatar, name, and current work status. Flossie's response demonstrated several key capabilities of agentic AI:

• Autonomous tool chaining: It proactively spun up a Codex thread to build the widget
• Cross-app collaboration: When Alex asked it to check Slack to see if any colleagues were working on something similar, Flossie immediately identified that Brian, Gad, and Nelson were already working on related widgets
• Agentic social actions: It could pull the user into the relevant Slack thread with those colleagues, and even draft a reply tagging the relevant people — waiting for the user's approval before sending

Throughout the interaction, DOT behaved more like a proactive intern who fills in the gaps than a passive tool. Alex ultimately chose to "let Flossie keep working" and check back at the results when he returned to his computer — which is exactly the core scenario for an always-on agent: tasks can keep progressing even after you've stepped away.

The "Codex thread" mentioned in the demo refers to OpenAI's Codex cloud sandbox execution environment. Unlike ordinary code completion that only outputs code as text, the Codex sandbox can actually run code inside an isolated cloud container, install dependencies, read and write files, and return execution results — including build artifacts and error logs — back to the agent. This enables DOT to complete a full iterative loop of "generate code → run → fix errors → run again," without requiring any local development environment setup on the user's end. Prototyping an iOS home screen widget involves the SwiftUI framework and the WidgetKit API, which carry a meaningful development learning curve for humans. DOT's ability to run this as a background async task that continues after the user walks away is a textbook example of sandbox execution combined with agent scheduling.

What This Means

DOT's arrival on mobile marks another step forward in OpenAI's push toward agentic AI. Putting a proactive agent on your phone means AI assistants are evolving from "a tool you use at your desk" into "an executor that goes everywhere with you." The ability to autonomously orchestrate tasks across calendars, email, Slack, and coding environments is what truly sets it apart from an ordinary chatbot.

Of course, this comes with an obvious trade-off: the more powerful the agent and the more systems it's connected to, the more data access and trust users must hand over. The balance between DOT's utility and the degree of privacy openness it requires will be a key consideration for users in practice.

For those who want to try it now, the barrier to entry is as low as it gets — just open the ChatGPT app and tap "Meet DOT" in the top-left corner to get started.