Software as toys

I've decided to write more frequently, documenting what I've been thinking and exploring, especially ideas and thoughts that keep surfacing. I think it'll help me understand myself better and think more clearly. Let's see where this goes.

Software as toys#

I build things with interesting new concepts to play with. When I finish building, I feel like a kid again, playing and sharing my toys with my friends in a classroom, and then shelve it if it feels boring. If my classmates like my toy, I can sell it to better sustain myself, if they don't like it, it's perfectly fine, I've had the fun I worked for.

I really like this concept of software. It feels pure, it's curiosity-driven and not market-driven. When you're building, you don't have to worry about "what the market wants", you worry about "do i think this is a cool/fun concept?".

It's only possible today with agentic coding. I built my new toy in a week, and the Adults (App Review) are checking if it follows the rules. I'll be sharing it soon!

Being curious today is a blessing#

Learning becomes so much more enjoyable because you get to ask a lot of questions and get answers a lot faster. In the past, if I stuck on an abstract text, I'd have to either spend a ton of time and energy grinding through it, or find someone knowledgeable on the topic who cared enough to help me out. And then if you're unlucky you'll have to deal with people thinking you are stupid. Not fun. But that's no longer the case.

My recent intellectual challenges were understanding the concepts in The Little Book of Deep Learning, and I think I can explain most of them simply now.

Investing in self-education#

We pay tuition fees in the multiples of 1000s before we graduate college, why shouldn't you do the same after you graduate? I justified spending $100 to acquire a 24/7 smart-ass tutor, and I think you should probably spend at least $20.

Human as boot loaders? Will the next generation of intelligence kill us?#

This has been making me uncomfortable for the past few days.

In Singularity is Near, the author argues that evolution works in a compounding way, with each generation of intelligence developing on top of the previous generation, "it creates a capability and then uses that capability to evolve the next stage". Atoms build molecules, molecules build amino acids and proteins, proteins built brains, brains built machines, and maybe we can merge with machines and transcend biological constraints. In this interpretation, machines will eventually expand human consciousness throughout the universe, it's more anthropic.

image.png

But an alternative interpretation is to view humans as boot loaders for Super Intelligence. A boot loader is a specialized piece of software that initializes a computer's hardware and loads the operating system into memory when the system is powered on or restarted. Once the boot loader has successfully performed its tasks, it hands over full control of the computer's hardware and resources to the operating system kernel. In that case we will live under the mercy of clankers.

It sounds really depressing, the technology that excites me might replace us in the end. But maybe that's just one possibility, and we can do much better rooting for and building the alternative future where we colonize the universe.

What's the most important thing after singularity?#

If anyone's wildest dream can come true, what's the most important thing after? Will it be education?

Coding changed again, and it's still changing#

I still code a lot, especially since I started working as a software engineer after graduating.

I think coding changed again after this generation of frontier coding models, Opus 4.5 and GPT-5.2-Codex.

In the last generation of coding agents, I still use IDEs a lot because I have to read the results of agents, I especially have to look out for excessive defensive logics that drastically decrease the quality of code (slop code). Besides this I have to read and maintain lots of different documents. My Cursor usage is basically 0 for the past few months.

But more recently, I found myself seldom opening Cursor or any other IDE and rely heavily on terminals and CLI agents. I often just open 7 Warp tabs, with two lazygit tabs for version control of different projects, and five agent tabs. The main reason I believe is that the quality of the code has gone up, and they just "get it" most of the time.

I don't have enough RAM to clone the repo several times so each agent works in different git checkouts, so I'll be the main orchestrator of the agents and make sure they don't work on the same subset of features/code at a time, which makes me the main blocker of development speed.

The recent observations are that:

1. For frontend, humans are the main blocker to productivity, because we have to manually click through the buttons to see if something is going to work. Automating testing is getting more and more important, and this is a skill that can be developed. I personally haven't discovered a good solution to close the iterative loop for agentic frontend coding.

2. Agent orchestration is going to be the next big thing, i.e. an organizational way of having Agents work with each other, eliminating the excessive human intervention in the loop. I feel like a lot of the explorations are currently being done here.

3. Terminal based coding is the current meta, but when orchestration comes out, UI with better information manipulation features will come back. Because orchestration means the work we hand off to AI has more complexity, more complexity means a good UI will be very helpful, so terminal-based coding won't last forever. Software engineering is still in a transitional phase.

4. vim is still a useful skill if you don't want to open an IDE to read files.

5. Knowing what's possible and matters becomes more and more important. If you know what's possible, you get to test your concepts. If it matters, well you should double down.