How I Work

I’ve always joked that the difference between genius and stupidity is roughly 24 months. My theory is that, given the same information, people often reach similar conclusions; some just get there much faster. I put myself somewhere in the middle, so I spend a lot of time looking for different inputs.

As I improve over time, I want a timestamped record of what I believed and why. I do this because I want to check my priors constantly.

I spend a lot of time reviewing my decisions against the information I actually had at that moment in time. When the outcome is good, were there better options I didn’t consider? What pitfalls did I get lucky to avoid (moral luck!)? When the outcome is bad, could I reasonably have done something differently with what I knew then? Would I still make the same decision with what I knew then?

I want to understand the risks I took, whether I took them with my eyes open, and what I should carry into the next loop. Hopefully by iterating on my decision-making, I can get a bit closer to the genius end of the spectrum.

I want to spend as much time as possible thinking about cool problems, spending time with awesome people, and building fun things.

This is the kit and plumbing I use to turn those inputs into outputs. The basic loop is capture, consolidate, analyze, and ship. Handwritten notes, meeting transcripts, and stray thoughts land in one place, where I can revisit them, look for patterns, and turn them into something useful.

I want to make capture as free as possible. If something has a cost, it often doesn’t get paid, and when ideas or thoughts disappear into the ether, something fun, exciting, or novel might get lost.

I’ve been a huge pen-and-paper person because writing things down physically helps me retain the information. I carry an A6 and a pen with me everywhere, from meetings to ballgames to poker rooms to the golf course. It also removes the digital-distraction element of technology and keeps me comfortable with my own sense of boredom.

The only downside of this kind of note-taking? It’s hard to search the records. I’ve been experimenting with the reMarkable Paper Pro Move as a way to get a better handwritten-to-digitally-indexable flow.

I use Granola to capture meeting notes (both full transcripts and summaries) and effectively run a regression across them to see which thoughts I keep repeating. The output lives nicely inside my Obsidian vault.

If I don’t have access to other mediums, I quickly jot things down in Todoist on my phone; it’s the easiest thing to reach when I’m trying to get something down fast.

Thoughts go on paper when I’m thinking, on my iPhone when I’m moving, and are captured automatically when I’m in a meeting.

Importantly, in the pursuit of low cost, none of this is organized at capture time. Sorting happens when I’m at my desk; that lets me review the notes and keeps them from landing in the wrong folder.

For me, the best future-proofing is keeping things searchable and in one place. Everything ends up in one Obsidian vault as Markdown. That’s the whole architecture: files on disk that any editor, script, or model can read, with no export step and no vendor between me and my notes.

Meeting transcripts arrive through Granola sync, handwritten notes from the reMarkable arrive through an OCR tool, and the daily Todone note ties together what I’ve done.

The vault syncs across all of my devices with Obsidian Sync and is mirrored via Dropbox on purpose, to prevent single points of failure and to make sure I can pick up any device without spending a huge amount of time on-ramping back into my workflow.

Amusingly, any new tool has to live in the vault or feed into it. It’s a great test for killing apps I’m just not going to end up using.

Once the notes are plain text, the models can work over them: what did this person actually ask for, what’s the follow-up, what does this week’s set of meetings add up to, and the standing backlog of analysis questions that never fit inside a call. ChatGPT and Claude both run on their top plans, on purpose, because they’re differently wrong and I want to see the disagreement. I also like Perplexity, mostly because I like how they tag search sources and I’ve liked the people there.

AI models read and synthesize; they never write or rewrite the source. None of my transcripts, ink, or anything else is ever edited by an LLM, and anything a model produces is a separate file that points back at the original source it read.

• Independent dual review. Anything code-shaped is bootstrapped with one assistant and reviewed by both before it merges
• Provider-agnostic. Pipelines swap or arbitrate between models rather than depend on one vendor
• Privacy first. Real notes and transcripts never leave the vault or enter a repo; only tooling is versioned
• Deterministic when it matters, probabilistic otherwise. Sync jobs are plain scripts with no LLM in the write path, but some tasks should live with, and cite, their revealed uncertainty

I’m a big believer in shipping early, often, and continuously. Shipping is a muscle, and if you don’t use it, you lose it. There’s also a nice hedonic bump from closing the loop.

Tasks get closed in Todoist. My writing lives on my personal website, in a newsletter, or on someone else’s blog (with some machinations to perhaps do more video and audio).

Prototypes go up on Shots.SebastianPark.com so people can click into them instead of just hearing me blabber on about it.

Ideally, every day ends with a Todone note. Shoutout to Nicole Aptekar, who put me onto this over a dozen years ago. Todone notes are not a plan but a record of completion; it’s all about what actually got finished. It’s both an honest metric for output and a way to remind yourself, when you’re in the middle of many projects, that you’ve actually done cool things. Give it a try! You’d be surprised how many things you’ve actually done.

I’m still living the circa-COVID-era RTX 3080, 10th-gen Intel Core i7 (8C/16T), 64 GB of RAM life. Windows box with WSL (h/t to Brian Shih): every project lives in a Linux home directory and I do my best to keep everything segregated. Windows is still the nuts for the gaming, the streaming, and the writing, but old vim habits die hard and I like Linux for the engineering-adjacent work.

One confession: my desktop is basically always on. I love that from any café anywhere in the world I can, via Tailscale, Sunshine, and Moonlight, get access to my home network and the devices on it. My camera/audio chain exists because I do a lot of calls and used to do a lot of recording. I highly recommend upgrading to a real camera and a real mic; it’s the best way to upgrade your system.

My 14" MacBook Pro is my everywhere-in-the-city machine: cafés, other people’s offices, a borrowed desk for the afternoon, Betaworks. It’s an M1 Max with 32 GB, which is still more than enough for writing, research, and most code. Anything that wants the GPU or the WSL environment isn’t done here; it’s done on the desktop through a Moonlight session, which means the laptop never needs to carry a second copy of anything heavy.

Obsidian and iTerm do most of the heavy lifting. All of it rides in a Rains Rolltop alongside an Anker 165W battery pack.

Maybe the most boomer thing I do: I still believe deep thinking has to happen on paper. Not as an aesthetic choice either; it’s just that paper lacks a feed, a tab, or anything else that could distract me. I have so many notebooks, scaling from A2 to A6 depending on whether it’s paper spread across a desk (A2), meetings where I’m paying close attention and can’t record (A4), or something that fits neatly in my pocket (A6).

The reMarkable Paper Pro Move is the newest thing in the kit and it’s on probation. The test it has to pass is simple: pages sync into Obsidian as PDFs, and the rmocr pipeline below turns the ink into searchable text while keeping the original. If a scribble from a train this year is findable in four years, it stays. I doubt the paper will ever leave.

I bring the MBP on most trips, but once I’m out and about in a city, I’ll often leave it behind and carry just the iPad. It has the benefit of fitting into my Osprey Daylite Sling (a crossbody bag that lives inside my Aer and functions a lot like an aircraft carrier and fighter jet). The iPad Pro with the Magic Keyboard is a great writing surface on the go, the iPhone 17 Pro has an amazing camera for capture, and a small notebook is always great for handling both sketches and quick thoughts. As always, the point is to take a note, answer a follow-up, and keep moving along without having to open a laptop or get gripped by the tendrils of technological distraction.

In terms of travel, I typically use the Away Bigger Carry-on for most trips, only switching to the Monos small carry-on for Europe, where overhead-bin enforcement is less forgiving. The Anker 165W pack keeps the whole kit alive throughout a long day without having to worry about power.

There are a bunch of gaps inside an ordinary day. The subway to a meeting, a cab across town, walking between two things, standing in a line (one of my least favorite things). These may be one-hand-only spaces. Here, the phone just works best. I also have a bunch of wearables, but really, it’s the Oura Ring as a record for sleep (I sleep pretty well, so I like the positive reinforcement of a “good job!”) and an Apple Watch on occasion when I want that type of tech on my wrist (golf comes to mind).

That said, an analog watch is great for when I’m taking meetings and don’t want to look at a digital screen but want to keep track of time. Messaging is split by who’s on the other end: KakaoTalk, WhatsApp, Telegram, iMessage.

Both frontier subscriptions on purpose. Code gets bootstrapped with Codex and reviewed independently by both Codex and Claude before it merges. Open problem: syncing context so work lives in both.