Sunny

← All weeks

What I’ve been reading (W42, 2025)

Data design, more AI thoughts and what's new in Postgres database and React Native

  1. Your data model is your destiny https://notes.mtb.xyz/p/your-data-model-is-your-destiny

The structure of your data defines the structure of your product and, over time, your company. A model that mirrors the real world of your problem space lets you express complexity cleanly and compound value with every new feature. The paradox is that this choice happens when you know the least about your market, but that’s also why it’s so powerful when you get it right. Once you commit, every decision - UX, infrastructure, APIs, even org design—flows downstream from it. Competitors can’t just “copy this feature.” They’d have to rebuild the substrate. For example -

  • Toast models restaurants as systems of menu items, modifiers, and kitchen routes — not as generic SKUs. Because those entities reflect how kitchens actually work, Toast can do native order routing, prep-time orchestration, and ingredient-level insights. What looks like “just a POS” to others is actually a simulation of the kitchen.
  • Figma is built around a shared, real-time canvas — a live object, not a saved file. This flipped the core abstraction of design tools from static documents to dynamic collaboration. Adobe couldn’t bolt that on; it would have meant rewriting their desktop-centric foundations.
  • Rippling made the employee record the centre of gravity. HR, IT, payroll, and finance are just different projections of the same object. Every new module inherits context, permissions, and automations by default. Competitors built in silos can only integrate via brittle syncs; Rippling compounds coherence.
  1. LLMs are terrible at asking questions. They just make a bunch of assumptions and brute-force something based on those guesses. Good human developers always pause to ask before making big changes or when they’re unsure. But LLMs? They keep trying to make it work until they hit a wall - and then they just keep banging their head against it. Sure, you can overengineer your prompt to try to get them to ask more questions - but it’s very likely they still won’t https://kix.dev/two-things-llm-coding-agents-are-still-bad-at

  2. [Following up on 2] But AI can be useful in other ways
  3. The biggest Postgres 18 difference is how it does I/O operations. It introduces a new io_method setting that gives users more control over how disk I/O is handled. There are two options: worker, the new default, which uses dedicated background processes for all I/O; and io_uring, which uses Linux’s io_uring asynchronous I/O interface to perform disk reads without blocking. The hope is that this change, especially with io_uring, will bring noticeably better I/O performance https://planetscale.com/blog/benchmarking-postgres-17-vs-18, https://pganalyze.com/blog/postgres-18-async-io

  4. React Native is making the new architecture the default from version 0.82 onwards https://reactnative.dev/blog/2025/10/08/react-native-0.82; the cool things in the new architecture include (all resulting in improved performance, reliability, and developer experience)
    • C++ based native module system with synchronous access
    • Direct JavaScript Interface (JSI) – no async bridge
    • Strongly-typed contracts via Codegen
    • Lazy (on-demand) native module loading
    • Cross-platform C++ renderer with immutable UI tree
    • Multi-threaded, interruptible rendering for better UI responsiveness
    • Well-defined event loop on JavaScript thread
    • Supports React 18 features: transitions, Suspense, automatic batching
    • Instant synchronous layout reads (e.g., tooltips, element positioning)
    • Backward compatibility layer for old libraries
    • Faster startup and improved error handling/debugging