Somewhere in the last few years, documents got demoted.

The tools at the centre of knowledge work are now chat products. The conversation is the main event. The document, the thing you're actually judged on, became an "artifact": a preview pane bolted to the side of a chat, regenerated top to bottom every time you ask for a change, forgotten when the thread scrolls away.

That's backwards. Nobody's work is a chat log. The work is the strategy paper, the board pack, the policy set, the client report. Documents are where thinking becomes concrete and decisions become durable. In the hub-and-spoke world of the chat tools, the document is a spoke. It should be the hub.

The old guard doesn't solve this either. Word and Google Docs take documents seriously, but they take them one at a time: a single document, a blank page, and AI sprinkled over an architecture designed decades before any of this existed. Ask the built-in assistant for help and it can see the document you have open. The research it should draw on, the transcript from yesterday's call, the four other documents that need to move when this one does: invisible.

And that's the real gap, because knowledge work is almost never one document. It's a folder. When I led an ISO 27001 certification, the "document" was two dozen policies that all had to agree with each other; one decision in a security review could ripple through five of them. When I do strategy work, one body of research feeds a working paper, an executive summary and a deck outline, and the moment the thinking shifts, every one of them is wrong until someone fixes them by hand. The tools see none of this. They see whatever is in the window.

Programmers already got the better deal. Tools like Cursor treat the project as the unit of work: an agent reads the whole folder, proposes changes as diffs, and never touches a line you didn't approve. I've spent years watching what that does to the pace and quality of building software. The unfair part is that this way of working stayed locked inside code editors, guarded by terminals and git, while the people who live in documents got a chatbot and a text box.

So here is what we think a first-class document surface looks like.

The project is the workspace. Your folder of real files, on your machine, opened at whatever level the work needs. Not an import, not a workspace you migrate into and can't leave.

The AI works across the whole project, not the open tab. Draft in one document with the research in reach. Change a decision once and carry it through everything it touches.

Every change is a proposal. An edit you can read, scoped to what you asked, shown as a diff before it lands. Trust in AI doesn't come from accuracy claims; it comes from being able to see exactly what changed and say no.

And documents can be alive. A report bound to its sources doesn't need rebuilding every Monday. It needs an agent on a schedule, updating what's safe to update and holding anything that changes the story for a human. Sometimes its most valuable output is "no change recommended".

None of this can be retrofitted. It isn't a feature you add to a word processor or a plugin you bolt onto a chat app; it's the architecture underneath. That's why we're building Abstract new, from the ground up, AI-native at the core: documents first, spreadsheets and decks to follow.

We're building it for the people who already know there's more available. You use Claude or ChatGPT every day. You've felt the ceiling. The models are capable of far more than the surfaces they've been given, and you shouldn't need to become a programmer to get it.

Documents are the work. It's time the tools believed that.