Codex to Dots is the short version of how OpenAI built its always-on personal assistant. Dots, OpenAI’s entry in this year’s big consumer AI race, did not start life as an assistant at all. It grew out of Codex, the coding agent OpenAI built for software engineers. That is the story Alexander Embiricos, who leads OpenAI’s enterprise product team, told the audience at TechCrunch Disrupt this month, in an interview TechCrunch published on 26 October 2026.
His main point is simple. Most of what Dots does was already possible with the Codex app, which OpenAI released at the start of the year. What changed was the packaging. “A lot of that isn’t building some new piece of tech into the product,” Embiricos said. “It’s how you frame it to the user and how it feels.”
This guide traces the Codex to Dots journey step by step and date by date, using OpenAI’s own launch posts. It then covers what Embiricos said about proactive research, privacy, permissions and payments, how Dots compares with Codex and ChatGPT Work, what it costs, and what the Codex to Dots shift means for businesses weighing one up. We first covered the product before launch, when it was reportedly called “o”, in OpenAI’s Upcoming Always-On Assistant Is Reportedly Named ‘o’.
Table of contents
- From Codex to Dots: The Family Tree
- Codex to Dots: Why Codex Could Already Do Most of It
- Proactive Research: The Codex to Dots Feature That Thinks Ahead
- Privacy and Permissions After Codex to Dots
- Payments, Agents and the Limits of Codex to Dots
- Codex to Dots Compared: Codex, ChatGPT Work or Dots?
- What Dots Costs After the Codex to Dots Repackaging
- How the Codex to Dots Bet Looks in the Assistant Race
- What the Codex to Dots Shift Means for Businesses
- Our Verdict on OpenAI’s Codex to Dots Strategy
- Codex to Dots FAQs
- References
From Codex to Dots: The Family Tree
Embiricos is well placed to tell the Codex to Dots story. In a December 2025 appearance on Lenny’s Podcast he was introduced as the Codex product lead. He joined OpenAI in 2024, when the company acquired Multi, the screen-sharing start-up he co-founded. He now leads OpenAI’s enterprise product team, and he was the person OpenAI sent to Disrupt to explain the Codex to Dots story.
The short version
OpenAI launched Codex in April 2025 as a coding agent. Over the following 17 months it gained a desktop app, background work, memory, plugins and computer use. OpenAI then wrapped that same agent technology into products for non-developers: workspace agents in April 2026, ChatGPT Work in July, and finally Dots on 29 September at DevDay. That is the whole Codex to Dots arc in one paragraph.
TechCrunch describes this as “repackaging the functionality of the company’s models into something more accessible, a path it was already on as it spun ChatGPT Work out of the agentic coding tools it had built for software engineers.”
The timeline in OpenAI’s own words
Every step below comes from an OpenAI launch post. Read in order, they show Codex to Dots happening one feature at a time.
| Date | Launch | What it added on the road to Dots |
|---|---|---|
| April 2025 | Codex | A coding agent for developers |
| 2 Feb 2026 | Codex app for macOS | Many agents in parallel, skills, and Automations that run on a schedule |
| 16 Apr 2026 | Codex for (almost) everything | Background computer use, 90+ new plugins, memory, proactive suggestions |
| 22 Apr 2026 | Workspace agents | “Codex-powered agents for teams” that run in the cloud and in Slack |
| 9 Jul 2026 | ChatGPT Work | “Codex technology built-in” for everyday work tasks |
| 29 Sep 2026 | Dots | One named assistant with its own cloud computer |
The gaps between launches
The Codex to Dots pace was steady. Counting calendar days between each launch date in the table:
Bars are scaled to the longest gap: 73 is 89% of 82, 78 is 95% and 6 is 7%. In total, the Codex to Dots run took 239 days, from the Codex app to the assistant.
The clues OpenAI left behind
You can still see the Codex to Dots roots in OpenAI’s own pages. Its ChatGPT Work guides for sales, finance and operations teams sit at web addresses ending in “use-codex”, such as “how-sales-teams-use-codex”. A guide first dated 23 April, now titled “Everyday work with ChatGPT Work”, carries a note that its webinar “references an earlier Codex experience”. The coding agent was renamed and repackaged for everyone else, and Codex to Dots is the latest step on that road.
Codex to Dots: Why Codex Could Already Do Most of It
Embiricos was frank about this part of Codex to Dots. TechCrunch reports that he “points out that you could achieve much of dots’ functionality with OpenAI’s Codex app”, but that doing so “requires more work and hand-holding.”
The Codex app was built for long-running work
OpenAI’s 2 February post described the Codex app as “a command center for agents”. It let developers run several agents in parallel on projects that “can span hours, days, or weeks”. It added skills, which bundle instructions and scripts so Codex can do work beyond code. It also added Automations, which “let Codex work in the background on an automatic schedule”. OpenAI’s demo had Codex build a racing game using more than 7 million tokens from a single prompt.
Every one of those pieces survives the move from Codex to Dots: background work, long tasks, tools and a schedule. What the Codex app lacked was a friendly front door for people who do not write code.
“Codex for (almost) everything”
The 16 April update moved Codex closer still. OpenAI said Codex could now “operate your computer alongside you”, with background computer use on the Mac. It released more than 90 new plugins, a preview of memory, and the ability to “schedule future work for itself and wake up automatically to continue on a long-term task, potentially across days or weeks.”
It also added proactive suggestions. In OpenAI’s words, Codex could now “suggest how to start your work day or where to pick up on a previous project”, for example by finding open Google Docs comments that needed attention. That is the seed of proactive research, the Codex to Dots feature we look at below.
Leaving the computer on all night
Embiricos told one story that sums up the hand-holding problem. He recalled celebrating when his wife started leaving her computer on all night so that her agents could keep working. Agents that run on your own machine need that machine to stay awake.
Codex to Dots removes that step. Each dot gets its own cloud computer and browser, so the assistant keeps working after you close your laptop. That single change, moving the agent from your desk to OpenAI’s servers, is much of what turns a power-user coding tool into an assistant that ordinary people can use. It is the biggest practical difference in the move from Codex to Dots.
One entity, one conversation
The other Codex to Dots change was framing. “As we were working on the product, we realized if we could frame this as, it’s an entity and it’s a single conversation, not many different conversations… it starts to become intuitive,” Embiricos said.
The Codex app organises work into separate threads, one per task. Dots presents one named character that holds all your context. OpenAI’s launch post makes the same point: “You can keep bringing it new tasks and ideas without having to juggle separate threads or direct every step.”
Mascots and the name
That framing explains the cute avatars. TechCrunch opened its interview by asking why so many companies are giving AI assistants mascots. Embiricos called it “convergent evolution” across the industry. In the Codex to Dots shift, a face and a name make the agent feel like one helper rather than a list of chats.
The name itself is a play on the negative space inside OpenAI’s logo. Embiricos said it was meant “to feel like just something that could eventually be ubiquitous.” Asked when he learned that Elon Musk owns the dot.com web address, he said: “I found out very late.” We covered that episode in The Internet Is Convinced Elon Musk’s xAI Trolled OpenAI’s ‘Dots’ Launch.
How fast Codex grew beyond code
OpenAI’s own usage figures show why the Codex to Dots repackaging made sense. Codex was already being used well outside software.
Bars are scaled to 5 million: 3 million is 60% and 1 million is 20%. OpenAI gave “more than 3 million developers” in April and “more than 5 million people” and “more than 1 million” non-software users in July.
Proactive Research: The Codex to Dots Feature That Thinks Ahead
The core idea behind Codex to Dots is an assistant that does not wait to be asked. Embiricos described the goal as an assistant “that’s sort of thinking about you and thinking how to help you.” TechCrunch notes the business side plainly: for OpenAI, a persistent agent means more token usage.
What proactive research does
OpenAI calls the feature “proactive research”. Embiricos explained it this way: “On a cadence, it will think about what it knows about you, what is the conversation that it’s had, and then what have you given it access to that it can think about, and then it will go and research, ‘What could I do to be more helpful to you?’ And then, basically, it will propose that.”
OpenAI’s launch post adds an important limit. When you are not actively working with it, your dot uses your connected apps “with tools that are restricted to be read-only, which means that they can’t send messages, change app content, or control your browser or computer.”
The three-minute warning
The feature is still rough. In TechCrunch’s testing, Dots flagged calendar conflicts and upcoming deadlines, but “not necessarily in the most timely fashion.”
Embiricos had his own example from early development. “The first time it told me about one of those [conflicts]… it told me three minutes before the thing,” he said. “It’s like, ‘Hey, this thing is coming up. You might want to do something about it.’ I only have three minutes!” He expects the timing to improve.
The outage story
His best example came from a user. He described an engineer at a drone delivery start-up who said that by the time they noticed an outage, their dot “had already noticed the outage was happening, had already analyzed the outage, and had come to me with useful information for how we could resolve the outage.”
That is Codex to Dots at its best: watching in the background and arriving with a diagnosis, not just an alert. It is also very close to what a Codex automation could already do for a developer, which is Embiricos’s point.
What OpenAI says it does inside the company
OpenAI’s launch post lists similar internal uses. “A bug appears in Slack, and dots immediately start investigating.” It also gives a developer example in which a dot watches customer feedback, builds and tests small fixes, and brings complete pull requests to review “with attached videos showing the changes.” After the move from Codex to Dots, the assistant is still, in part, a coding tool.
Privacy and Permissions After Codex to Dots
Moving from Codex to Dots also moves an agent from your code into your personal life. Any assistant like this needs a lot of personal data to be useful, and that raises the obvious question: how much are you handing to a software company? TechCrunch put it to Embiricos directly, asking whether Dots collects the kind of data Meta’s Muse does.
Yes, it learns about you
Embiricos answered “Yes”, and pointed to proactive research. Your dot thinks about what it knows about you on a regular cadence and goes looking for ways to help. TechCrunch notes that Muse has been criticised for building profiles of your family and friends, and that Hark, a start-up building its own assistant, stresses that it does not collect data to fuel an advertising business the way OpenAI and Meta do.
We looked at those promises in AI Agent Makers Are Promising Privacy — Will They Deliver?.
What OpenAI’s controls cover
According to OpenAI, Dots has the same privacy controls as ChatGPT. You can turn off model training on personal plans, and OpenAI does not use content from Business, Enterprise or Edu workspaces to train by default. OpenAI also says it does not train directly on proactive research or on a dot’s notes to itself.
What TechCrunch found in testing
The controls are not granular, and TechCrunch says the app can be pushy. “For example, it refuses to accept read-only access to cloud drives.” That sits awkwardly beside OpenAI’s description of read-only proactive research: the background tools are restricted, but the connection you grant can still be broader than you want.
The table below sets OpenAI’s description beside what the interview and testing found.
| Area | What OpenAI says | What the interview and testing found |
|---|---|---|
| Background work | Proactive research uses read-only tools | The app would not accept read-only access to cloud drives |
| Controls | Same privacy controls as ChatGPT, plus Custom Rules | Not granular yet; Embiricos wants “super, super, super granular controls” |
| Approvals | Built-in rules decide when to act and when to ask | Launch defaults were so conservative that users complained |
| Training | Opt-out on personal plans; no direct training on proactive research | Not tested by TechCrunch |
“Super, super, super granular controls”
Embiricos accepted the criticism. “What I think we need to do is basically build out super, super, super granular controls,” he said. He also pointed to the model’s own judgement as the other half of the answer: a smarter assistant should need fewer hard rules.
The permission-prompt problem
The opposite complaint came at launch. “We had really conservative defaults over what the agent would do, and it would ask for permission a lot,” Embiricos said. “I remember knowing that we were going to see that feedback, but thinking, ‘This is okay, we can tune that over time.'”
That trade-off came along with Codex to Dots, because Codex users were developers used to approving changes. Ask too often and people stop reading the prompts and click yes to everything. Ask too rarely and the assistant acts without consent. OpenAI’s own safeguards include auto-review of risky actions, an Activity View for background work, and a rule that sensitive tasks such as changing a password “always stay with you.” For companies, this is as much a cybersecurity question as a privacy one, because a dot holds live connections to email, calendars and files.
Payments, Agents and the Limits of Codex to Dots
The last part of the interview looked ahead. If every person gets an assistant like this, can the world’s businesses, websites and payment networks cope?
Talks with payment processors
Embiricos hinted at the next step. “At OpenAI… we’re not always moving the fastest in terms of shipping the next ability for the agent to work with the world,” he said. “When we do, we’re doing it really closely in collaboration with the rest of the ecosystem. I probably can’t say too much here, but we’re deep in lots of conversations with payment processors.”
An assistant that can spend money would be a big step for OpenAI and a big ask for users. TechCrunch’s reporter put it bluntly: giving an assistant access to an email inbox already feels like a big ask, but a bank account is “off limits. For now.”
Nobody wants to meet your agent
There is also a social limit. Businesses are not always equipped to deal with AI agents, and people often do not like dealing with them. Embiricos is one of those people. He has a human assistant, because OpenAI’s corporate partners want to deal with a person. “I personally never intend to have a meeting with someone else’s agent,” he said.
Specialist dots for companies
OpenAI is also taking Codex to Dots inside companies with a second kind of dot. Its launch post describes “specialist dots” with their own identity, credentials and access, set up by a company to handle defined jobs. OpenAI says it has tested them internally in procurement, invoice processing, email marketing, customer support and commercial contracting, and is starting with focused enterprise pilots. It is also working with Microsoft to manage specialist dots through Microsoft Agent 365.
Codex to Dots Compared: Codex, ChatGPT Work or Dots?
Because the Codex to Dots products share so much technology, it is easy to mix them up. Here is how they differ, based on OpenAI’s launch posts.
| Feature | Codex app | ChatGPT Work | Dots |
|---|---|---|---|
| Built for | Developers | Knowledge workers | Anyone on an eligible plan |
| How work is organised | Separate threads per task | Tasks and projects | One named dot, one ongoing conversation |
| Where it runs | Your computer, CLI, IDE or cloud | Web, mobile and desktop | Its own cloud computer and browser |
| Background work | Automations on a schedule | Scheduled Tasks | Always on, with proactive research |
| Where you reach it | The app, terminal or editor | ChatGPT | ChatGPT, Slack, Teams and voice calls |
| Model at launch | Codex models | GPT-5.6 | GPT-6 Astra |
When the extra layer helps
If you are a developer who already runs Codex automations, Codex to Dots adds less for you than it does for everyone else. Embiricos more or less said so. A reviewer on The AI Maker newsletter found a good use that bridges the two. Their most-used routine runs every day at 8 a.m.: the dot reads all of the previous day’s Codex conversations and says which tasks need attention. The same reviewer watched Dots launch three separate Codex tasks on its own, with prompts the dot had written.
When it does not
If you need precise control over each step, separate Codex threads are clearer. Dots trades some of that control for convenience, because it decides what to look at and when.
They work together
The products are not walled off from each other. You can ask your dot to start or manage tasks in Codex or ChatGPT Work, so Codex to Dots is not a one-way move: your dot can act as a front desk for the older tools.
What Dots Costs After the Codex to Dots Repackaging
Codex to Dots did not make the technology cheap. TechCrunch says it is available in the ChatGPT desktop and mobile apps to subscribers on the Pro tier or higher, with plans starting at $100 a month.
Plans and markets
According to OpenAI, Dots is rolling out to Pro and Business Premium users in eligible markets. Enterprise workspaces, including Edu and Healthcare, can try a beta if an admin turns it on. You create your first dot in the ChatGPT desktop app or a desktop browser, and can message it on mobile after setup.
The Pro price ladder
OpenAI’s Pro plans come in three tiers. All of them include Dots, so the cheapest way to try Codex to Dots for yourself is the $100 tier.
Bars are scaled to $500: $100 is 20% and $200 is 40%.
How usage is counted
Your first dot is included in a Pro or Business Premium plan at no extra cost. OpenAI says conversations with your dot do not count toward your ChatGPT usage limits, but tasks it starts in Codex or ChatGPT Work do. OpenAI plans to let users add more dots later, and to scale each one’s speed or monthly workload.
That pricing model matters. An assistant that keeps thinking about you in the background burns tokens even when you are not typing, which is why it sits at the top of OpenAI’s price list.
How the Codex to Dots Bet Looks in the Assistant Race
TechCrunch frames the always-on AI assistant as “this year’s answer to the question of how people will actually use deep learning models.” OpenAI is not alone.
The main rivals
| Assistant | Company | What stands out |
|---|---|---|
| Dots | OpenAI | Built on Codex; own cloud computer; Pro plans and up |
| Muse | Meta | Heavy marketing; criticised for profiles of friends and family |
| Instinct | Start-up | Became “an overnight sensation”, in TechCrunch’s words |
| Hark | Start-up | Says it does not collect data for advertising |
For more on the head-to-head, see OpenAI’s New Agent Is a Shot at Meta — but Can It Compete With Free? and Instinct Was the Buzziest AI Agent Around — Can It Survive Muse?.
OpenAI’s edge
OpenAI’s advantage from Codex to Dots is that its assistant sits on a stack that millions of people already use for work. Codex had more than 5 million weekly users by July, and ChatGPT Work is being rolled out to entire companies; OpenAI said on 22 October that Uber is bringing it to more than 30,000 employees.
OpenAI’s weak spots
Reviews of Codex to Dots have been mixed. In a hands-on test for WIRED, which we covered in OpenAI Wants Its New Agent to Run Your Life. In One Test, It Said “I Love You”, Dots produced useful shopping research but also some odd moments. Embiricos himself admits that proactive timing and privacy controls need work.
What the Codex to Dots Shift Means for Businesses
For most companies, the Codex to Dots shift means a product to pilot carefully, not to roll out to everyone on day one. Here is a practical way to think about it.
Start with the jobs Codex already proved
The strongest tasks after Codex to Dots are the ones that grew up inside Codex: watching for problems, summarising activity, preparing drafts and fixes for review. OpenAI’s own examples, such as investigating a bug reported in Slack or noticing a missed invoice, fit that pattern. Begin there.
Map the access before you connect anything
Any assistant is only as safe as the apps it can reach. Before connecting email, calendars or cloud drives, list what each connection allows. TechCrunch’s finding that Dots would not accept read-only access to cloud drives is a reason to check scopes yourself rather than assume them. Our guide to AI agents covers how to plan permissions for autonomous tools.
Set rules, then review them
Use Custom Rules to require approval for anything that sends, buys or deletes, and check the Activity View regularly in the first weeks. Expect to loosen the rules over time, as Embiricos suggests, but start strict.
Budget for the token bill
Proactive research runs whether or not you are typing. At $100 a month per Pro seat or a Business Premium plan, plus usage on any Codex or ChatGPT Work tasks your dot starts, costs can add up across a team. Measure the hours saved against that bill before expanding.
Keep a human front door
Embiricos’s own habit is a useful guide. Even the person who explained Codex to Dots on stage keeps a human assistant for outside contacts, because partners want to deal with a person. Use your dot behind the scenes, and keep people in the conversations that matter to customers.
Our Verdict on OpenAI's Codex to Dots Strategy
Codex to Dots is less a new invention than a new wrapper. The hard parts, including long-running agents, background schedules, memory, plugins and computer use, were built for Codex and shipped to developers first. What Embiricos’s team added was a single character, a single conversation, a cloud computer that never sleeps, and a habit of offering help before you ask.
That is a real product decision, not a trivial one. Most people will never set up a Codex automation, but many might talk to a named assistant that remembers them. That is the bet behind Codex to Dots. The open questions are the ones Embiricos named himself: better timing, finer privacy controls, and whether the world is ready for agents that spend money. Until those are answered, treat Dots as a capable preview of where Codex to Dots is heading, and keep your bank details to yourself.
Codex to Dots FAQs
What is Dots?
Dots is OpenAI’s always-on personal assistant, launched on 29 September 2026, and the latest step in the Codex to Dots line. Each dot runs on GPT-6 Astra with its own cloud computer and browser, connects to more than 4,000 apps through plugins, and keeps working in the background.
How did OpenAI turn its coding tools into Dots?
Codex to Dots means Dots reuses the agent technology OpenAI built for Codex. The Codex app added background Automations in February, and an April update added computer use, memory and proactive suggestions. Workspace agents and ChatGPT Work then brought that technology to non-developers, and Dots packaged it as one assistant. That is Codex to Dots in short.
Who is Alexander Embiricos?
He leads OpenAI’s enterprise product team and was previously the Codex product lead. He joined OpenAI when it acquired Multi, the start-up he co-founded, in 2024.
Can Codex do what Dots does?
Mostly, according to Embiricos, which is the point of Codex to Dots. He said much of what Dots does is possible with the Codex app, but it takes more work and hand-holding, such as keeping your own computer running overnight.
How much does Dots cost?
Dots comes with ChatGPT Pro plans, which start at $100 a month, and with Business Premium. Enterprise workspaces can try a beta. Your first dot is included at no extra cost.
Is Dots private?
Dots uses ChatGPT’s privacy controls, and OpenAI says proactive research uses read-only tools and is not trained on directly. TechCrunch found the controls are not yet granular, and Embiricos said OpenAI needs to build much finer ones.
Will Dots be able to make payments?
Not yet. Embiricos said OpenAI is “deep in lots of conversations with payment processors”, but gave no date for a dot that can spend money.
References
How OpenAI turned its coding tools into dots, an always-on personal assistant (TechCrunch)
Introducing the Codex app (OpenAI)
Codex for (almost) everything (OpenAI)
Introducing workspace agents in ChatGPT (OpenAI)
ChatGPT is now a partner for your most ambitious work (OpenAI)
Everyday work with ChatGPT Work (OpenAI Academy)
Uber brings ChatGPT Work to all employees with OpenAI (OpenAI)
Why humans are AI’s biggest bottleneck, with Alexander Embiricos (Lenny’s Newsletter)
How to Use OpenAI Dots as Your AI Chief of Staff (The AI Maker)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
