OpenAI’s dots brings coding-agent ideas into ChatGPT

OpenAI is pushing its agent strategy beyond coding with dots, a personal assistant designed to keep working in the background rather than wait for each prompt. According to TechCrunch, the company’s enterprise product lead Alexander Embiricos said dots grew out of work on OpenAI’s coding tools and aims to make agent behavior feel more natural for everyday work.

For freelancers, creators and small teams, the important shift is not a new chat window. It is the move toward an AI helper that can see calendars, email, files and local context, then suggest action before a user asks.

From coding workflows to one ongoing assistant

OpenAI introduced dots at its developer day in late September, according to TechCrunch. Embiricos described the product as part of a broader convergence among AI companies trying to turn large models into assistant-like entities rather than isolated prompt-and-response tools.

The assistant is available inside the ChatGPT desktop and mobile apps for subscribers on the Pro tier or higher, with plans starting at $100 a month, TechCrunch reported. Once connected to a user’s calendar, email, computer and other tools, dots is meant to help with work across those systems.

The product’s roots are in OpenAI’s coding stack. Embiricos told TechCrunch that much of what dots does could already be achieved with OpenAI’s Codex app, but that approach requires more manual setup and supervision. The product challenge, as he framed it, is less about inventing entirely new model capabilities and more about presenting them in a form that feels useful and intuitive.

TechCrunch also reports that ChatGPT Work came from agentic coding tools OpenAI had built for software engineers. In that context, dots looks like another step in turning developer-focused automation into a broader workplace assistant.

What dots does before you ask

The core idea is persistence. Instead of waiting for a user to start a task, dots is designed to keep track of relevant signals and propose help. Embiricos described it to TechCrunch as an assistant that is thinking about the user and ways to be useful.

In TechCrunch’s testing, that proactive behavior showed promise but was not fully reliable. The assistant could flag calendar conflicts and upcoming deadlines, but the timing was not always ideal. Embiricos gave an example from early development in which his dot warned him about a conflict only minutes before it mattered.

The kinds of work TechCrunch described include:

• spotting calendar conflicts and deadlines;
• continuing tasks in the background;
• analyzing information from connected tools;
• proposing next steps based on what it can access.

Embiricos also cited an example involving an engineer at a drone delivery startup, where the assistant had reportedly noticed an outage, analyzed it and surfaced useful information before the engineer had seen the issue.

For small businesses, that suggests a near-term role as a triage layer: useful for reminders, early warnings and context gathering, but not something to trust blindly with time-sensitive decisions.

The privacy trade-off is hard to avoid

Dots depends on access. To be useful, it needs permission to inspect personal and work systems such as email, calendars and files. That creates the same tension facing many AI agents: the more context they have, the more helpful they may become, but the more sensitive data users are handing over.

Embiricos told TechCrunch that OpenAI built a feature into dots called proactive research. On a recurring basis, the assistant considers what it knows from conversations and connected resources, then looks for ways it might help and proposes them.

OpenAI says dots uses the same privacy controls as ChatGPT. Users can turn off model training so their conversations are not used to improve OpenAI’s models, and they can control which resources the assistant accesses and how. But TechCrunch noted that those controls are not yet granular and said the app can be insistent, including refusing read-only access to cloud drives.

Embiricos acknowledged that OpenAI needs more fine-grained controls. He also said the product launched with conservative defaults and asked for permission often, even though the company expected some criticism for that approach.

That puts dots in the same practical debate as other agent products: access management is becoming as important as model quality. We have seen related questions around agent logins and hidden passwords, as well as financial institutions restricting how agents use customer credentials in banking contexts.

The outside world may be the bigger bottleneck

TechCrunch raised another problem for always-on assistants: scale. Agents may be convenient for individual users, but businesses, websites and other people may not want to interact with them.

Embiricos said he has a human assistant because OpenAI’s corporate partners want to deal with a person. He also said he personally does not intend to meet with someone else’s agent.

The payment question is especially sensitive. Embiricos said OpenAI is moving carefully when giving agents new ways to act in the world and is in conversations with payment processors, though he said he could not share much more.

For now, dots should be viewed as a proactive assistant for work context, not a fully autonomous business operator. The useful test for creators and small teams is simple: would you be comfortable giving this tool access to the data needed for the task, and can you review its suggestions before anything important happens?