Sunny

← All weeks

What I’ve been reading (W18, 2025)

Building useful software, what being an owner feels like, how technology gets cheaper, vibe trading, designing for AI

  1. A lot of software these days is built to be sellable, and that’s what comes out of the software industrial complex. Palantir does it differently: they build “thick” software that earns its keep by sitting on the factory floor with users, watching what breaks, and fixing the real problems that make people fall back to Excel. By embedding with customers and iterating in the mess of actual operations, much like Elon Musk’s teams rebuilding their own manufacturing stack, they aim for software that’s truly valuable, not just marketable https://www.youtube.com/watch?v=jB13kCmWT2k
  2. If you really care about something, you get hyper involved in the details. If it’s your thing, you notice the chair is too far to the left or the font is a bit too bubbly. If you’ve never owned anything, this isn’t always clear. But if you’ve gone through life being an owner— an owner of your thoughts, your work, your destiny— nothing annoys you more than when committees ruin a tiny detail of the thing you obsess over and don’t get why it’s important https://x.com/KTmBoyle/status/1917940204456878145
  3. Learning curve improvements, where costs fall by some constant percentage for every cumulative doubling of production volume, are a major driver of technologies getting more affordable, but they can only take you so far. If improvements are difficult to find, costs might stay high and production volumes stay low even at high learning rates https://www.construction-physics.com/p/where-are-my-damn-learning-curves
  4. Trading is very bizarre these days, markets are moving based on headlines and vibes. Gold is up, the dollar is down, and bonds are all over the place. The traditional relationships that once gave investors a sense of order - bonds as a hedge, the dollar as a signal, gold as safety - are eroding, if not fully broken https://kyla.substack.com/p/the-united-states-of-amazon
  5. Uncertainty in the models, slow responses, and non-deterministic outcomes makes designing AI applications hard. You have to prototype to productise, prune possibilities fast, and keep mechanics ahead of aesthetics so users still sense the value to succeed here. https://lukew.com/ff/entry.asp?2095