Why AI needs a better word processor
Chat boxes are a terrible place to write. Here's what we learned building Docs — a word processor where agents show their work, and you stay in control of every edit.
The ThinkBench Team Ask anyone how they write with AI today and you'll hear the same routine: draft in a chat window, copy the result into a document, notice everything that's wrong, paste it back into the chat, and ask for changes. Repeat until you stop caring.
That routine isn't writing. It's laundering text between two tools that were never designed for each other. We built Docs because we think the fix isn't a smarter chatbot — it's a better word processor.
The chat box is the wrong shape for prose
A chat conversation is linear and append-only. A document is neither. When your only interface to AI is a message thread, three things break immediately:
- You lose the diff. The model rewrites your paragraph wholesale, and it's on you to spot what actually changed — and what quietly got worse.
- You lose the scope. You wanted one sentence tightened; you got a whole new draft with a different voice.
- You lose the history. Version four lives in the chat, version five in your document, and version six in a second chat you opened when the first one got too long.
None of these are model problems. They're interface problems, and interface problems are solved by software.
What writing with AI should feel like
Our rule for Docs was simple: every AI capability has to arrive through a mechanism a word processor already understands — comments, suggestions, versions, selections. Nothing lands in your document that you didn't see coming.
Edits arrive as reviewable diffs
When an agent edits your draft, the changes show up the way a colleague's suggestions would: change by change, each one acceptable or rejectable on its own. You stay the editor. The agent stays a contributor.
Selections scope the work
Highlight a paragraph and hand just that block to the agent. It edits in place — it can't wander off and restyle your conclusion because it never saw your conclusion.
Diffusion fills the gap you shape
Our favorite: write two sentences, drag a box between them, and size it to the number of lines you want. A diffusion model fills the space, connecting your ideas in place — no prompt, no chat, just the shape of the paragraph you were missing.
The best AI interface for writing isn't a conversation about your document. It's your document.
This applies to more than words
The same philosophy runs through Viz, our data-visualization tool. Because every chart is a
readable block expression — source |> group_by |> pivot |> sum — an agent can compose or
refine a visualization the same way you would, and you can read exactly what it did. No black boxes, in prose
or in charts.
Try it yourself
Docs and Viz are both included in every ThinkBench plan, starting at $20 a month, and the open core is free to self-host. If you've ever pasted a paragraph into a chat window and winced at what came back, we built this for you.