Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableTechCrunch→ original

OpenAI Turns Codex into dots, a Proactive Assistant That Works on Its Own: From $100 a Month

OpenAI has released dots, an AI assistant that runs in the background and offers suggestions without waiting for your prompt. For example, it notices a conflict in your calendar or a system failure and brings a ready-made analysis. In essence, dots repackages work from Codex, the coding agent, for ordinary people. Access is available in ChatGPT on the Pro plan and above, starting at $100 a month. For product managers, the lesson is this: sometimes the presentation matters more than the new technology.

AI-processed from TechCrunch; edited by Hamidun News
OpenAI Turns Codex into dots, a Proactive Assistant That Works on Its Own: From $100 a Month
Source: TechCrunch. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTechCrunch ↗
Editor-in-Chief
PlanChatGPT Pro and above, starting at $100 a month, on desktop and mobile
Key metric #2The first warning about a calendar conflict arrived 3 minutes before the event
Key metric #3dots does not accept read-only access to cloud drives

OpenAI has released dots, a personal AI assistant that runs in the background and offers help on its own instead of waiting for you to type a request. Picture an employee who sees two overlapping meetings in your calendar or an approaching deadline in your email and tells you about it unprompted. Or an engineer whose production system fails: before he notices it, the assistant has already analyzed the causes and offered suggestions for fixing it.

dots is available in the desktop and mobile versions of ChatGPT to subscribers on the Pro plan and above, and those plans start at $100 a month. The product was shown at the end of September at OpenAI's developer conference, and Alexander Embiricos, head of the company's enterprise product group, told TechCrunch Disrupt where it came from.

Not a New Technology, but a New Package

The key is presentation. According to Embiricos, most of dots' capabilities are already available in Codex, the programming app OpenAI released earlier this year. But Codex requires more manual work: setting things up, explaining, steering. He himself was pleased when his wife started leaving her computer on overnight so her agents could keep working.

So the team decided to package the models' existing abilities in a way that makes them easier to use. The idea is that the assistant is a single "entity" and one continuous conversation, rather than a set of scattered chats. That makes it intuitive. The company has taken a similar path before, when it split ChatGPT Work out of its agent tools for programmers. Even the name is a play on the empty space inside the OpenAI logo. The domain dot.com, as it turned out, belongs to Elon Musk, and Embiricos learned about this "very late."

There is no shortage of competitors: the startup Instinct became a hit almost instantly, and Meta is spending heavily on marketing for its assistant Muse.

Proactivity and Its Cost

The point of a constantly running agent is that you do not have to ask it. For OpenAI this means higher token consumption; for the user, an assistant that thinks on its own about how to be useful. The mechanism is called proactive research: at set intervals the agent recalls what it knows about you, which conversations took place and what access it has been granted, then looks for ways it could help and proposes them.

In TechCrunch's tests, the ability to look ahead is noticeable and often useful, but it is far from ideal. dots warns about calendar conflicts and deadlines, but not always in time. Embiricos said that in early development, his dot flagged a conflict only 3 minutes before the event. He expects the response time to improve over time.

Privacy, Scale and Money

A constant assistant needs a lot of personal data. Muse has been criticized for building profiles of family and friends, which are useful to an assistant but could also be valuable to Meta's advertising business. Hark, a startup building its own assistant, explicitly states that it does not collect data for advertising, unlike OpenAI and Meta. OpenAI says dots has the same privacy settings as ChatGPT: you can turn off training on your conversations and manage which computer resources the agent can access.

But, according to the journalist's observations, the settings are not very flexible, and the app can be intrusive: for example, it does not accept read-only access to cloud drives. Embiricos acknowledges that "very, very granular" permissions are needed. At launch the product was criticized for overly cautious settings and frequent permission requests, but the team plans to adjust this over time.

The question of scale remains. Companies do not always want to communicate with other companies' agents, and they are not always ready for it. Embiricos keeps a human-facing assistant because OpenAI's partners want to talk to a person, and he "never plans to meet someone else's agent." At the same time, OpenAI is negotiating with payment processors. If the assistant could spend the user's money, that would be a potentially lucrative step for the company, but a very serious request to people: even access to email is hard for many to grant, and a bank account, as the author notes, is currently off limits.

The practical takeaway for developers and product managers: agent features already present in engineering tools can become a mass-market product if the presentation and access permissions are carefully designed. But flexible permissions, timely suggestions and trust remain the weakest points for now.

Story

Key Questions Answered

What are the key details on Not a New Technology, but a New Package?

The key is presentation. According to Embiricos, most of dots' capabilities are already available in Codex, the programming app OpenAI released earlier this year. But Codex requires more manual work: setting things up, explaining, steering. He himself was pleased when his wife started leaving her computer on overnight so her agents could keep working. So the team decided to package the models' existing abilities in a way that makes them easier to use.

What are the key details on Proactivity and Its Cost?

The point of a constantly running agent is that you do not have to ask it. For OpenAI this means higher token consumption; for the user, an assistant that thinks on its own about how to be useful. The mechanism is called proactive research: at set intervals the agent recalls what it knows about you, which conversations took place and what access it has been granted, then looks for ways it could help and proposes them.

What are the key details on Privacy, Scale and Money?

A constant assistant needs a lot of personal data. Muse has been criticized for building profiles of family and friends, which are useful to an assistant but could also be valuable to Meta's advertising business. Hark, a startup building its own assistant, explicitly states that it does not collect data for advertising, unlike OpenAI and Meta. OpenAI says dots has the same privacy settings as ChatGPT: you can turn off training on your conversations and manage which computer resources the agent can access.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Hands-On AI Engineering · Hamidun Academy

Build & Ship Production AI Agents Without Boilerplate Code

Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.

Start 72h Free Trial →Vibe-Coding Intensive
What do you think?
Loading comments…