Hamilton Ulmer

Tools for Data Analysis & Deep Work

The writing tool of my dreams; a meditation
Feb 14, 2023

I think it’s a genuine challenge to make a writing tool that exploits the magic of the web in the right way. Google Docs and Notion both get a lot of things right (and in different ways), but I’m still craving something that goes toward the grain of the writing’s Grand Mess.

LLMs will clearly make tools better. But I do feel that an ounce of UI is worth a pound of AI. The jobs to be done of written communication are still going to exist.

The writing tool of my dreams has these features:

  1. It’s collaboration-first, & not just comments-off-to-the-side. Any good online writing tool should design for collaboration and commenting first, not last (which is the norm). Google Docs is probably the best I’ve seen in this regard, but the comments themselves become an off-to-the-side part of the experience instead of a central element. A lot of docs end up with “comment graveyards” – unaddressed points where you have no idea what the endpoint of a conversation meant for the doc.
  2. It keeps space for ideation and annotation. Collaboration comes in two forms: with others, and with future versions of yourself. A unified collaborative workflow should consider both of these cases. Yet very few writing tools make it easy to attach notes or metadata to paragraphs as you’re writing. So it’s hard to annotate what you’re doing so you can get things right as you go.
  3. It exploits the web platform for all its strengths. The web is the most revolutionary communication medium since the printing press. Yet most tools treat the web itself as an implementation detail and not a target. It feels like a huge waste of potential that we’re still authoring solely for a paper experience – especially technical writers. Let me effortlessly embed interactivity into my writing!
  4. It honors the idea graph. The linearity of most writing tools only really makes sense from a presentational point of view; but when writing, you’re basically creating a graph of insights. Being able to reshape your document and figure out the flow != skeuomorphic typewriter experience. Additionally, if I have a collection of documents, I may have an explicit graph – pages as nodes in a graph – and an implicit graph – ideas encapsulated within the pages. I think this is the structure people want to uncover.
  5. It’s fundamentally spatially-oriented. I think “skeumorphic” does make sense if you think about the physical approach to writing – a bunch of paper you can re-arrange spatially. This is why I think the future-best writing tool is probably an infinite canvas, or at least has some spatial affordances.
  6. It feels like a “professional tool”. A good writing tool is a “professional tool”, much like Pro Tools, Figma, or Photoshop. Professional writing tools are made for deep work, designed from the correct holistic conceptual model of the task and workflow, and require a bit of learning to really unlock their power. Tana gets some of the features right; so does Notion.
  7. It uses LLMs not as a hat-trick, but to augment the actual deep-writing task. We’re obviously entering a period where people stick LLMs into existing writing tools. Incredibly useful writing aid. But it’s still basically a robot-using-typewriter experience. An LLM should go way deeper than curing writer’s block or summarizing. I want it to help me structure arguments and improve the flow of my writing, not just auto-complete an idea in a mundane way. In a way, it’s not dissimilar to what I want LLMs to do with data warehouses & transformation DAGs. I want to give a system a schema, queries, and usage data, and then uncover the true structure of the business. A writing system built into a document store (like Notion or Google Docs) Should do something similar.

I wouldn’t count on Google Docs getting this right; Notion probably could, given their ability to actually move in new directions.