Why I built writtten
I am not as afraid that AI will take my job as I am afraid it will take my mental ability and skills.
AI tools are awesome for productivity. But in some places where they made things easier, they might have made them too easy.
For product managers and similar roles, the place where it got too easy is writing. Product requirement documents, communications, tasks, emails: most of this text is now generated with AI, usually through some form of chat. The chat interface is as native as it gets, we’ve been interacting with it for years. But it makes it feel too easy to delegate all the writing that is (was?) a big chunk of the core work.
And moving from a document to a chat changes more than the interface. Writing in a document is how I think via text, how I exercise my expression and voice. AI does some heavy lifting when working on documents. But if I am only steering and never lifting anything myself, won’t my thinking and writing muscles atrophy?
Stopping AI use altogether is not really wise, seeing all the benefits it brings. What I wanted was to “regress” back to the document view and have AI by my side while I work on it. I didn’t find a tool that would do exactly what I wanted, so I built one.
What writtten is
A document interface where you write everything yourself, and the AI only highlights substantive issues in the text:
“this contradicts the success metric you set two sections ago,”
“this claim has nothing supporting it,”
“you never addressed the risk you promised to cover.”
It deliberately stays away from grammar and style fixes. The AI is always nearby to detect, highlight, and support, but it never edits anything for you. You are the sole author of the text.
It also tries to pick its moments: while you are still getting the words down it stays out of the way, and it becomes more opinionated as the text settles and you start revising.
I don’t expect writtten to pull people away from working on docs with agents. Even this page was written in writtten, and I asked an agent to check some things after I finished. But that’s the whole point: I didn’t prompt it to write anything. I focused on the writing myself and benefited from a few notes I got along the way, without damaging the flow.
What to expect
writtten is early, and there is a lot I still want to work on (the quality of the notes above all). Expect some of them to miss the point, and expect some things worth catching to slip by. Nevertheless, I find it already quite useful.
A few practical things: it’s bring-your-own-key (your key, your data, everything stays local in your browser), and there’s a built-in “See it in action” example if you want a feel for it before setting anything up. The project is also open source, and you can run it locally.
If you try it, share your feedback (especially the roasting kind). The best place for it is a GitHub issue.