Sunny

← All weeks

What I’ve been reading (W2, 2026)

Ozempic changing food habits, stages of AI knowledge work, why writing matters, AI coding tactics, and programming on the subway

  1. Ozempic, that drug people take to lose weight, is changing how they spend money on food. In households where at least one person is on this drug, total food spending drops about 5%. Savoury snacks fall roughly 10%, and fast-food spending drops around 8%. The only thing that actually goes up is yoghurt. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5073929
  2. Most knowledge work with AI still looks like riding a bike with a bunch of loose parts - context is fragmented, and outputs aren’t easily verifiable, so we have to glue everything together. Once context gets consolidated and work becomes verifiable, people move from pedalling to driving. And eventually, a lot of that work becomes self-driving. https://x.com/ivanhzhao/status/2003192654545539400
  3. Writing lets you escape your mind’s default setting. It forces you to tidy mental clutter and articulate ideas with a level of context and coherence your head alone can’t manage. Writing expands your working memory - you can be more precise, even more brilliant, on paper than in conversation. It also exposes what you don’t know, the gaps and weak spots you can’t see when ideas stay trapped in your head. And it surfaces the ugly, uncomfortable, or unrealistic parts of your thinking. If a thought, feeling, or belief keeps looping, write it out. https://kupajo.com/write-to-escape-your-default-setting/
  4. AI coding tactics: scope management and sharing relevant context is everything; try different models when outputs are lousy; use coding agents at every step of SDLC - writing tests, fixing bugs, opening PRs; use git worktrees for isolated AI experiments; steer your AI assistant by providing style guides, examples, and rules files; let your CI/CD, linters, and code review bots catch mistakes automatically. https://addyosmani.com/blog/ai-coding-workflow/
  5. In the coming years, companies that update processes to improve the non-code chokepoints will reap the largest rewards - like code review bandwidth, testing/validation, release confidence, and product decisions. The software delivery process should be updated to be accessible to agents: deterministic tests, fast local runs, one-command dev environments, docs, golden datasets, perf harnesses. https://x.com/pauldix/status/2006423514446749965
  6. Programming on the subway - some inspiration to code during your commute https://www.scd31.com/posts/programming-on-the-subway