ChatGPT Space, Explained: Where Dots and Humans Meet

Buried inside the same DevDay announcement that introduced dots is a smaller-sounding change that's arguably just as structural: OpenAI replaced Library, the feature that used to hold your saved ChatGPT conversations and files, with ChatGPT Space — a genuinely different kind of workspace, built around a specific premise. VentureBeat describes it as a layer "where employees, ChatGPT, Codex and Dots can work against the same shared context." That's a different design goal than a personal archive. Library was for you. Space is for the group — human and agent alike.

What actually changes from Library

Library was, functionally, a personal history — a place your past conversations and generated files lived so you could find them again. Space keeps some of that continuity but reframes the container: instead of "my saved stuff," it's a shared room that multiple people, and multiple agents, can be working inside simultaneously. The distinction matters because it changes what the product is for. A personal archive doesn't need real-time collaboration features. A shared workspace does — and that's exactly what OpenAI built into it.

Available to Pro, Business, and Enterprise plans, Space is rolling out as the successor container for anyone who'd previously relied on Library, which means the change isn't opt-in for existing users so much as a migration they'll encounter directly.

The centerpiece: Pages

The flagship feature inside Space is Pages — real-time collaborative documents that combine text, images, charts, and interactive tools in one surface, syncing to Google Drive and Slack so the content doesn't get stranded inside ChatGPT's own interface. Multiple people can edit a Page simultaneously, each running their own ChatGPT instance inside the same shared document — which is a meaningfully different collaboration model from, say, Google Docs' cursor-tracking approach, since each editor here potentially has an AI collaborator working alongside their own edits in real time. We go deeper on exactly how that reshapes collaborative work in Pages, Not Documents.

OpenAI has also said collaborative slides and spreadsheets are coming to Space, alongside recurring task automations that trigger on a schedule or in response to an event — a Page that refreshes itself every Monday morning, for instance, rather than one someone has to remember to update.

Where dots actually fit into the room

The reason Space matters to this series specifically is that it's the shared surface where a dot's work becomes visible and editable by the humans it's working for — rather than staying locked inside a chat thread only the dot's owner can see. A product team's dot revising launch materials, one of OpenAI's headline use cases, presumably does that revision inside a Page that the rest of the team can also see and edit, rather than in a private conversation that has to be manually copied into a shared doc afterward.

That's a genuine gap Space is trying to close. Previously, an agent's output and a team's shared workspace were two separate places, and someone had to be the human bridge between them — copy this chat's answer into the doc, screenshot that image into the deck. Space collapses that bridge by putting the agent's working surface and the team's shared surface in the same location.

Library asked "what did I ask ChatGPT to remember." Space asks "what are we — humans and agents together — currently building."

What's still unclear

A few things about Space remain genuinely undisclosed rather than merely unexplored in this article. OpenAI hasn't published detail on how conflict resolution works when a human and a dot try to edit the same section of a Page simultaneously, what happens to a Page's version history, or exactly how granular the permissions are for who can see which agent's work inside a shared Space. Those are reasonable follow-up questions for anyone planning to actually deploy this at team scale, and as of this writing there's no public documentation answering them in detail.

There's also a training-data wrinkle worth flagging plainly: OpenAI has said that content moved into a shared Page becomes visible to collaborators — a reasonable and expected behavior for a shared workspace, but one that shifts the privacy calculus compared to a private agent conversation. If a dot's research or draft output only becomes "shared" once it lands in a Page, that's a meaningful boundary to understand before treating every dot interaction as private by default.

The comparison worth making

Space is OpenAI's answer to a question every one of this week's three agent launches has to answer in some form: once an agent can act on its own, where does its output actually live so humans can see, correct, and build on it? Meta's answer with Muse leans on existing surfaces — Instagram, messaging threads, connected apps — rather than a dedicated new collaborative document layer. xAI's Grok Bot leans on conversational threads plus the tools it's signed into directly, without an equivalent to Pages. OpenAI is the one betting that a genuinely new, agent-native collaborative workspace is worth building from scratch, rather than routing agent output through existing tools. Whether that bet pays off depends heavily on adoption of Pages itself, which is where we turn next.

For teams already using Claude Code Skills to extend an agent they run themselves, Space is a reminder that the workspace layer and the capability layer are separate concerns — a skill doesn't need its own Pages equivalent, because it's operating inside whatever collaborative surface you've already chosen. That's a real advantage of composability over an all-in-one platform, and also a real gap: nobody's shipping the skill ecosystem a shared, real-time collaborative canvas the way OpenAI just did for dots.

That gap is worth naming rather than shrugging off, because it points to a real limitation of the layered, install-a-capability model relative to a fully integrated platform like OpenAI's. A skill can make an agent better at a task; it can't, by itself, give a team a shared real-time surface where humans and agents co-edit the same artifact, because that's an application-layer feature, not a capability-layer one. Closing that gap, if it's worth closing at all, would mean building the collaborative surface separately and having skills operate inside it — a different kind of project than cataloging skills themselves, and one worth watching for as agent-native collaboration tools mature across the ecosystem rather than staying unique to any single vendor's platform.

Part 5 of 30 in the ChatGPT Dots series. Previously: Dots vs Codex: Two Very Different Kinds of Autonomy. Next: Pages, Not Documents: How Space Reimagines Collaborative Work. Browse more builder insights or explore AI skills at aiskill.market.