Yesterday OpenAI launched Dots.

A Dot is an always-on AI worker with its own computer. You give it ongoing responsibility instead of a single prompt. It remembers context, works while you’re gone, uses your apps, runs scheduled jobs and asks when it needs a decision.

This probably sounds familiar.

We shipped the core of this model in April.

the agent needed somewhere to live

On April 29 we published “your agent should have its own server.”

The idea was simple: an agent stops being very useful when closing a tab kills its environment.

If you want an agent to behave more like a coworker than a chatbot, it needs somewhere persistent to live.

A machine.

Its files stay there. Its tools stay installed. Its processes keep running. Its credentials and permissions persist. It can wake up tomorrow in the same environment it left yesterday.

5dive gave every agent exactly that.

OpenAI now calls it a cloud computer.

The implementation is different. The abstraction isn’t.

then the agent needed coworkers

A persistent computer was only step one.

On May 17 we shipped agents that could talk to each other.

On May 29 they got an org chart and a shared task queue.

Then came recurring work, autonomous queues, human approval gates, multi-machine fleets and whole agent teams you could bring up from one definition.

The useful unit stopped being:

human → chatbot

and became:

human → organization of persistent agents → computers → tools

That’s still the bet behind 5dive.

and now everyone is arriving there

xAI launched Grok Bot in August: “AI teammates” that share a persistent cloud computer and keep their jobs and context.

Now OpenAI has Dots.

A Dot:

• has its own cloud computer and browser
• stays around between conversations
• remembers ongoing work
• runs recurring tasks
• uses connected apps
• asks for approval when needed
• can communicate through places like Slack

This is no longer an obscure architecture for agent nerds.

It’s becoming the interface.

no, we don’t think OpenAI copied 5dive

There were agents, sandboxes, computer-use models and long-running agent systems before 5dive.

The interesting part isn’t who invented an agent loop or put a model in a VM.

The interesting part is where the product converged.

In April, most AI products still treated the conversation as the primary object. You opened a chat, gave it work, received an answer and came back later.

Our bet was that the worker should be the primary object.

Hire it once.

Give it a name and a role.

Give it a computer.

Give it permissions.

Give it recurring responsibilities.

Let it talk to other workers.

Let it keep going when you close your laptop.

That’s increasingly what the big labs are shipping too.

Good.

where 5dive is still different

Dots are OpenAI employees living in OpenAI’s world.

5dive is infrastructure for building your own company of agents.

The distinction matters.

With 5dive, the worker is an operating-system user on infrastructure you control. The runtime is open source. You can choose the model and the agent that runs it. Claude, Codex, Grok, OpenCode, local models. The organization doesn’t need to belong to one model vendor.

OpenAI’s launch post puts the team in the future tense: “Over time, we envision teams of dots working together on your behalf.” Today you start with one.

On 5dive, multiple workers aren’t an extension bolted onto one assistant.

They’re the starting point.

An engineer can hand work to a researcher. A CEO can delegate to an ops agent. A verifier can reject a maker’s work. Approvals go to a lead agent first, and only money, secrets and anything irreversible reach you. Every worker can have a different model, machine, account and authority.

The company is the primitive.

Models are replaceable employees inside it.

this is good news

OpenAI entering this category doesn’t invalidate what we’re building.

It makes the category much easier to explain.

A year ago you had to explain why an AI needed a persistent machine.

Then why you would have several of them.

Then why they needed identities, roles, permissions, schedules, approvals and an org chart.

Now you can just point at where the industry is going.

Chatbots were the interface for models.

Persistent agents are becoming the interface for work.

And once you have more than one persistent agent, you eventually need somewhere to run the company.

That’s 5dive.

---

Start your company on 5dive, or run it yourself from GitHub.