Sunny

← All weeks

What I’ve been reading (W12, 2026)

Understanding over abstraction, the limits of LLMs, OpenAI's side quests, and a guide to canned fish

  1. Trial and error will get you something that sort of works, but it leaves you with a system you don’t understand and can’t trust. Every abstraction you lean on without understanding just adds another unknown, and eventually you hit a point where nothing makes sense and progress stops. The frustrating part is that most of this isn’t hidden. You can read the code, follow what it actually does, and make it obvious. If you don’t, you’re not solving the problem, you’re avoiding it, and that’s how you end up stuck far longer than you needed to be https://blog.information-superhighway.net/on-the-need-for-understanding
  2. On GitHub’s recent downtime, all the outages were less about one dumb mistake and more about a big system quietly becoming fragile under growth, where saturation, hidden coupling, and under-tested failover paths turned ordinary changes into broad outages https://surfingcomplexity.blog/2026/03/12/quick-thoughts-on-github-ctos-post-on-availability/
    • Note to self/ - always plan for load honestly, avoid hidden abstractions that create knowledge gaps and keep configurations explicit and tested, not assumed
  3. Good writing isn’t just pattern and correctness, it’s about perspective, which is why LLMs struggle with it. They learn from averages and are trained to be safe and predictable, while real writers do the opposite, they break patterns, take risks, and write from their own lived experience https://www.theatlantic.com/technology/2026/03/ai-creative-writing/686418/
  4. OpenAI’s leadership is admitting they’ve been running too many side quests at once, from Sora to hardware to a browser, while Anthropic stayed focused on the things that actually drive value and proved it through enterprise and developer adoption. Now OpenAI is working to bring urgency and discipline back, hiring SaaS talent, tightening execution, and trying to convince investors they can still lead as the IPO window starts to close https://om.co/2026/03/17/openai-has-new-focus-on-the-ipo
  5. It’s hard not to think about building software now https://craigmod.com/essays/software_bonkers/
  6. A guide to eating canned fish https://www.youtube.com/watch?v=jIPmNmnRy7o