Tools for living - building tidepools

tldr: try tidepools - a daily journaling and task management app with a proactive AI coach.

When I left Google I had the sudden, unfamiliar experience of a surplus of time.

I had spent years working at big tech companies, managing complex cross-functional efforts, keeping people informed. A big part of my job was synthesizing information from different sources and then communicating it out. I lived in shared docs, bug trackers, metrics dashboards, and video conferences. Everything I captured and shared had to be as visible as possible. Notes on how decisions were made and who was responsible for what in meeting notes docs. Roadmaps in Google Docs. Time-sensitive communications in email. I spent a good part of my day managing all these different streams, making sure everything was captured and up to date so the collective intelligence of the team could respond nimbly as circumstances changed. It felt like a very specialized athletic performance.

When I left I lost all that scaffolding: I no longer had a packed schedule of meetings, no shared docs or bug trackers to keep up to date. My time was my own and, on both global and local scales, I had to decide how to use it.

It wasn't hard to find things to do: I tendered my resignation from Google in February 2020, and by April 2020 (my plans for a month in Italy pretty much shot) I was volunteering for an organization getting masks and gloves to healthcare workers. After that was a road trip, a retreat, a consulting gig. I even helped start a non-profit app for tracking wildfires.

Then a few things happened all at once: I lost a paper diary I'd been keeping for years, ChatGPT launched, and I started using Logseq for task management.

It wasn't obvious at the time that these things had anything to do with each other. ChatGPT was interesting but hallucinated a lot. The paper diary (a nearly full yellow Moleskine left in an airplane seatback) still hurts to think about. And Logseq, an outliner in a daily journal format with flexible to-do tracking, felt nearly perfect.

Slowly, though, everything seemed to come together: Logseq's flexible daily format encouraged me to write more and more. I started writing about my work instead of just adding and checking off to-dos. ChatGPT added tool calling, and I looked for something interesting to do with it. I wrote a simple CLI to-do app letting the user define their own ChatGPT functions to mutate the tree of to-dos (break this task up into sub-tasks, clean up this set of to-dos, etc.)

And one day in a Starbucks I created a toy project to learn the text editing framework lexical. This was before Claude Code, and lexical was too new to be in coding models' training data, so I was writing 90% of the code by hand, with occasional experiments using ChatGPT to write front-end code for me. I decided—why not?—to build a daily journaling app as an experiment.

I made a few decisions that differentiated it from Logseq. I didn't like being forced into an outline format, so I let pages be arbitrary Markdown. I didn't like how editing queries for to-dos in Logseq required writing code, so I added Excel-like formulas and a "find" function. Before too long, I was managing my development in the app itself.

The surface area for a really usable journaling app is huge. You need desktop and mobile clients. Server sync. Text editing is all edge cases. This all took time. Years! Friends asked when they could use it, but I hemmed and hawed. I made a list of people who offered to help kick the tires and did nothing with it. But I kept working.

Logseq had a feature that let you schedule recurring tasks, which I used to make sure tasks in important, non-urgent areas didn't get neglected. I thought it would be interesting to make that a little more flexible by putting AI in the loop. Suddenly I was getting daily suggestions from a model with access to my past journal entries and to-dos. At this point, my journal entries were long, exploratory. They covered projects, daily life, and all the emotions that came up during the day, so there was plenty of material for the AI to work with. I wondered what would happen if I asked it to generate higher-level insights and suggestions over a longer timeframe, so I added weekly review and planning workflows.

Eventually I had what I'm using today: a flexible tool for writing about and tracking my work, with a proactive coach that helps me reflect on a regular basis. I spend most of the day keeping a log of what I'm doing, what I'm struggling with, what I'm happy about, where I'm stuck. I've started to recognize what it feels like when I'm pushing through some unacknowledged inner resistance, and how easy it can be to stop doing that and inquire about what the resistance is, and what it wants. Procrastination is rarely an issue any more, but that's almost an afterthought. The value is in making my emotions about work first-class parts of my day, rather than irritating distractions.

At first I thought the coach would be most helpful in suggesting tasks at the start of the day. It’s true that this can be useful, if it has enough context about your work. But it turns out I've gotten the most value when it asks me probing questions, when I journal about something and it encourages me to dig into that area just a little bit more.

This is a relatively light-touch use of AI, in some ways: it's just generating some questions on a certain schedule. It is, mostly, still just a journaling app, where the value is created by you sitting down and writing.

But the impact these questions can have on you, when something resonates and causes you to see things differently, can be profound.

I built tidepools to solve a problem for me. The next step is figuring out how an AI journaling partner can help other people. If you like journaling, or you’ve wanted to try it, I'd love for you to try the app.