• Obsidian for note-taking has been a game-changer. The customizability it offers, mainly in terms of look and feel, and extensibility, from its rich plugin ecosystem, sets it apart for me. Notion, to me, falls short on the former and offers more features than I need. The modular nature of Obsidian makes it more like the Linux terminal. You compose your workflow from the modules that you require. The ability to export them into SVG, or high-quality PNGs is great, too. Many similar apps offer SVG export only with paid versions, a key differentiator for Obsidian. I've also been drawn to its local-first, and file-over-app philosophy. You're drawn to tools that align with your deeper self in some way, despite its flaws. You are more forgiving of the tool for its limitations if it "speaks" to you, either by giving you the space to channel your creativity, or through the values and principles its creators share. Obsidian checks both of those boxes for me. In the world of LLMs, where there's gonna be no dearth of personal software(any good programmer could spin up a note-taking app in a matter of days or weeks), it's gonna be important to find ones that have a core set of features that you care about the most and do them well, but that also offers the most extensibility and customizability. If your goal is to start writing more, any app will do, including the beautifully built Mac default, Pages. You find your tastes and preferences along the way. So don't get bogged down by choices.

• Excalidraw plugin in Obsidian has been an absolute gem. Going into 2024, I wanted to get better at expressing my ideas visually, even just for myself, if not for other people. And what's better than an infinite canvas to do that? The learning curve with such a tool is pretty low. You open a giant canvas and start drawing. At work, I've used LucidChart/LucidSpark for similar needs. Around mid-year, I started using Wacom One to draw on my Mac. Nothing fancy, but using a pencil-like tool for drawing is much better than dragging a mouse to draw curves. While all of this has helped me present my ideas visually at work, the most benefit has been personal. Looking back, I see a significant change in how I think. I think more visually now.. and hence expressing them that way has become easier.

• I pay to use icons from The Noun Project, which has also become a critical tool in my visual toolkit. I use it to augment my visuals in Excalidraw. I like to focus on conveying, or clarifying, an idea than crafting intricate drawings, so I find the appropriate icon from the noun project, customize the colors, download the SVG and import them into my canvas, and draw on top of them to connect the dots.

• Claude.ai has been my go-to AI chatbot. The only times I've used ChatGPT are when I'm at work, or when Claude is down.

• In my role, I don't get to code as much at work. Halfway through the year, I developed an itch to get my hands dirty writing some code. However my goal was to get familiar with coding LLMs and write tools that I'd personally find useful. So I started using Cursor with Claude and have not looked back since. I don't think I've even tried VS Code with Copilot(maybe for a days, or a week at most)! For someone who had not written any code in over a year, Cursor took me to be productive in a matter of hours. I created my blog. I created an Obsidian plugin - Serendipity. And I created some fancy data visualizations with it.

• That itch I talked about, about getting back into coding and getting better at visual communication, took me straight to learning D3.js and Svelte. I considered starting with React since I had used it a couple of years ago, but pretty quickly realized how heavy-handed it is, and how much you have to know to get going. I deliberately looked for something more intuitive and simple and Svelte, if those are your criteria, easily wins.

• NotebookLM, from Google, has been a relatively recent add, and I'm still exploring how to use it best for my needs. But it's quickly becoming one of my favorite and go-to tools when I'm learning something. One area it holds a lot of promise in is creating curated notebooks with high-quality content about specific topics, which can then be used for your notes, to build knowledge over time, and to write and publish blog posts. Like the article [[Long Context and Its Implications]] says, in a world where context length is virtually infinite, the curated content that we feed into the model is what differentiates good knowledge from bad ones. That's what differentiates experts and non-experts. I wish I had this in my grad school. Learning would have been so much more fun, especially the hard topics.

• I started exploring the following only a few weeks ago, so I'm not sure if it should make it into this year's list, but I'll mention it for completeness. It'll also act as a reminder for me to continue to invest in them next year. Evidence.dev with DuckDB.. While D3 is great for bespoke, artsy, and absolutely beautiful, vector visualizations(like this one), it's also quite some work for routine ones. Using something like DuckDB and plugging the data into Evidence is a pretty nice way to go about simpler visualization needs. I'm currently using this to pull my Reader, and Strava data from the year and create a look back, to derive insights about my reading and fitness activities.

Next year, while I want to continue to get better in the above areas, and optimize my workflows using the tools, I also want to invest in the following -

• LLMs like v0, to create beautiful, full-stack web apps. I'm sure this area is ripe for explosion and I'm looking forward to better models and tools. The part that's most frustrating to me about building web apps today is my lack of ability to craft beautiful things. I feel handicapped without a designer. I can tell good design from bad design, but can't come up with good design! I can build something functional, but they don't look good, and that's where I give up, because the UI/UX makes or breaks it for me. I hope to build useful and beautiful things next year.

• Local LLM models. I've dabbled with Ollama a bit this year, but for my current use cases, I feel like local models are not up to par with Claude Sonnet and Gemini Pro that back Claude.ai and NotebookLM. I'm sure we'll see huge progress in this space in 2025. Like the local-first philosophy that I alluded to before, I like the idea of using safer, local LLM models than uploading my content over the wire and into the hands of big corporations, especially as I start to use them more and more with intimate and private content.

• Build more malleable software

• Craft visual essays like Maggie Appleton, or write something as deep and as beautiful as Amelia Wattenberger