Sunny

← All weeks

What I’ve been reading (W17, 2026)

Notion reinventing itself as an AI execution layer, the loop-vs-loop shift, plus a security reminder and antirez on web bloat

  1. Notion went from docs to a place where AI workers execute on your data https://colossus.com/article/inside-notion
    • It started as a better Google Docs, but in 2022 the founders made a hard pivot and decided Notion is an AI company before the market caught up. And now Notion is one of only three of the most-used AI companies that didn’t start as AI-first https://x.com/Johnsjawn/status/2033248325185765410
    • This turned Notion from a place where work is stored into a system where work actually happens and this is not adding AI features. It is rebuilding the product into an execution layer powered by AI workers (See this launch https://vercel.com/blog/notion-workers-vercel-sandbox)
    • The reason it works is structural. Years of deeply organized company data plus a composable product make it possible for AI to operate directly on workflows
    • They also have a cool office!
  2. AI is no longer model vs model. It is loop vs loop https://kwokchain.com/2026/04/23/cursor-and-spacex-in-search-of-a-complete-loop
    • Models alone are not enough. The advantage comes from pairing models with products that generate real usage and feedback
    • The real system is a loop. The model powers the product. The product generates usage. The usage becomes data. The data improves the model
    • Coding is where this loop works best. It is structured, testable, and every failure is useful training signal
    • What takes multiple steps today becomes simpler over time. The messy workflow is exactly what trains the next version
    • Anthropic saw this early through coding agents. Others are now moving in the same direction
    • The Cursor and SpaceX deal fits this model. Cursor brings product and users. SpaceX brings compute. Together they close the loop.
  3. Good reminder from Sentry’s CTO (on row-level security) - even though isolation can be painful to maintain, especially with operations and data lakes, it’s still the right call for security and now, with AI workloads, that decision looks even more justified https://x.com/zeeg/status/2046251663909171489
  4. Antirez (the guy who created Redis) thinks web is a mess because big-company frameworks took over, simplicity got replaced by abstraction, and now we have bloated apps built by people who know the tools better than the basics https://news.ycombinator.com/item?id=47853663
  5. Google Cloud Run now lets you safely run code, use GPUs, handle background tasks, and store data - all in one place, making it easier to build and run AI apps https://cloud.google.com/blog/products/serverless/whats-new-for-cloud-run-at-next26, https://www.youtube.com/watch?v=AoisAy_LGpI @ernest