DISCLOSURE: If you buy through affiliate links, I may earn a small commission. (disclosures)
I'm now 6 weeks into my 12 week batch at Recurse Center. RC pushes you to learn generously and has a tradition of return statements where you reflect and share your learnings in the batch - so here I wanted to do a little early return statement to reflect on my time so far and where I want to go with the rest of the batch. (I'm big on regular reflections).
What I wanted coming in
First we should start off with my ideas for what I wanted to do coming in: What I Plan to Build at Recurse Center.
I had 3 buckets / areas I wanted to explore:
- Video games and simulations
- Web fundamentals
- Applied AI
What I've accomplished and learned so far
From my buckets:
- Video Games and simulations - Built a couple games like a Raylib + C# incremental clicker and AI Token Clicker with TS + Phaser. I've now got a better understanding of the underlying tools, patterns, and my own aptitude / affinity for them. I learned that I like the core systems but dislike all the polish and balancing that goes into making a game really fun. So I will keep building video games and simulations but keep them minimal and low fidelity so I get to do the things I like and stay away from the ones I don't.
- Web fundamentals - Arguably nothing here since I came in with the idea of looking at web internals. But I have shipped a couple webapps including rewriting my blog from C# to Rust and launching CloudSeed Rust (a fullstack Rust boilerplate). I learned that I really like building things I feel are useful and while learning for learning's sake is arguably useful, I prefer learning by doing instead. This often means building smth that incorporates the thing I want to learn vs just following a tutorial or reading up on theory (I like doing that but typically more in a JIT learning kind of way).
- Applied AI - I've spent a lot of time here though again not really going at it from a theory standpoint but more a practical standpoint. I've gotten my agentic AI engineering workflows dialed in balancing quality, speed, and learning and did a presentation and workshop on it at RC. I've also worked on building my own local-first AI agent orchestrator and thinking about how this kind of thing changes the future of software engineering. So far my takeaway is that I really like engineering with AI - in many ways it makes the possible feasible. I've always liked the building of systems and I've found paradigms to keep myself close to the system design even if I'm not writing the underlying code myself.
Outside of my buckets:
Things I've learned so far
So to distill a couple learnings from my time so far:
- I like applied knowledge, not so much theoretical knowledge. I learn best by doing and find that much more satisfying so leaning into small projects I want to build that incorporate the thing I want to learn seems to be the best fit for me.
- I enjoy quick, deep projects. I like going in, building the thing in 1-2 weeks, launching it, and moving on. I want to focus on the important bits and care much less for the polish / tweaking at the edges. This also means building projects that require little ongoing maintenance if at all possible. This also led to the demise of my idea to write a book during RC - it just felt like slog work and would take many weeks when I could just write the blog posts I enjoy and get on with my life.
- Following your curiosity is almost always the right choice. I do think there's a balance here where if you overindulge curiosity you never get anything done cause you're just chasing new idea after new idea. But I also think that if you have good systems in place to build a few things each week, keep solid prioritized lists of what you want to do, and consistently iterate on those things then a bit more serendipity is likely a good thing. I've learned over the years that no matter how much I distill my values into quantitative models, there's always a qual component of my curiosity / interest telling me to do smth and it's those things I'll typically enjoy more and put more effort into cause it feels fun vs work. That happened here with Rust and I'm really glad I pursued it. So I guess really have a balance - a set system that works for you, prioritized values to filter out the things that don't align and keep you focused, but then within those be open to flexibility and following the thing that's pulling you.
- AI requires a balanced approach. Finding a balance of quality, speed, and learning is the hard part. They each pull in different directions. This is the tradeoff dilemma and how far you should go in any one way depends on who you are, what you value, and the situation you find yourself in. It's these tradeoffs that make engineers, and generally human intelligence, valuable and it's this judgement / taste making that will remain valuable going forward. My current approach is to use a full vibe engineering workflow that mirrors that of a full software development lifecycle - I get to thoroughly explore an area, learn about it, think through possible solutions, and drive towards a system design I like. Then I work with the AIs to actually build it. I find this gives me similar quality to doing it myself much faster, plus I often do a couple more rounds of iteration because it's so fast to do so frequently the quality is HIGHER than I would've created myself. But the right balance / approach is likely going to be different for each person. For more on approach, see: 5 AI Coding Best Practices from a Google AI Director (That Actually Work).
What's next
We'll see where the batch takes me but some buckets I want to keep exploring and building in:
- High-level Rust - Rust is a great high level language but I think tools / docs for this are largely outweighed by the systems-level approaches. So I want to keep exploring in this area and building tools / examples to make this easier.
- AI Engineering - I personally believe this is the future and it's here. So I'm going to keep investing in my own workflows and grapple with what it means for my craft and profession. In particular how do I keep optimizing the balance of quality, speed, and learning. How do I stay in the loop, limit comprehension debt, and still improve on my old velocity by multiples?
- Web apps and tools - I'm going to keep building webapps and maybe some local tools for myself. These will likely scratch my own itch but I do hope to make them general enough that someone else could reasonably use / benefit from them as well. I have a habit of building, iterating on boilerplates so CloudSeed Rust will likely be the basis of the webapps and I may come out with some for CLIs / desktop / mobile apps depending on how many of those I make. I haven't done hypermedia in a while but am a big fan so might jump back in w HTMX or Datastar.
- Small games and simulations - I'm no longer interested in building a game polished enough for Steam but I am interested in spinning up atomic simulations on a given subject. I think AI can really help with the scaffolding here and I think it could be really useful for my essays if I could have some interactive demos or smth directly in the browser.
But who knows? I'm sure I'll get nerdsniped by smth new and unexpected and want to leave ample room for that.
Next
I've had a really good time at Recurse Center so far and am excited I have another 6 weeks to experiment and build ~full time with this community.
If you've been thinking about taking some time to explore / experiment with computers, I'd highly suggest giving them a look.
If you want to follow along with my builds and updates, you can find a link to my RSS feed below for ~weekly posts and my email list for a monthly roundup of what I've been up to. And if you're a Recurser, I post a checkin ~daily in Zulip.
If you liked this post you might also like: