Sunny

← All weeks

What I’ve been reading (W9, 2026)

Agentic engineering, hoarding working knowledge, Stripe at GDP scale, AGI's real constraints, and disaggregated databases

  1. Programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You’re spinning up AI agents, giving them tasks in English and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools , memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier “agentic engineering” feels very high right now. It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others, especially for tasks that are well-specified and where you can verify and test functionality. The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. https://x.com/karpathy/status/2026731645169185220
  2. A big part of the skill in building software is knowing what’s possible, what isn’t, and having at least a rough sense of how the possible things could be done. Questions like these can be broad or obscure. Can a web page run OCR in JavaScript alone? Can an iPhone app pair with a Bluetooth device even when it isn’t running? Can you process a 100GB JSON file in Python without loading it all into memory first? The more answers like this you carry around, especially ones you’ve seen working in real code, the more likely you are to spot ways to apply technology that others simply don’t see. And with coding agents, you only ever need to figure out a useful trick once. Document it with working code and an agent can recombine those pieces for you later. https://simonwillison.net/guides/agentic-engineering-patterns/hoard-things-you-know-how-to-do/ cc: @joy-ooo-back-on-18-nov @elliot @remus @han
  3. Last year, Stripe processed about $1.9 trillion in payments, roughly 1.6% of global GDP. It supports around 5 million businesses and is profitable. Zooming out, the market feels different now. Winners and losers are being decided faster, the gap between them is widening more quickly, and software products are in high demand https://assets.stripeassets.com/fzn2n1nzq965/3LlGw839Q6kUwxZlLZDtH6/75ddcbada4aa7743dd8ec7d0f9ca497e/Stripe-annual-letter-2025-desktop.pdf
  4. If AI starts doing most cognitive work reliably, labour stops being the binding constraint and compute and power take its place. In that world you don’t ask which SaaS company wins, you ask whether copper is mispriced, whether coal and grid capacity are mispriced, and whether hundreds of billions in knowledge exports are actually durable value or just expensive token generation waiting to be automated. There’s a wild question in here too: can AGI be achieved on 14nm chips with 10x the watts? If so, advanced chip fabrication matters way less than raw energy supply. And if AI is truly deflationary, how would we even know? https://danielgross.com/agitrades
  5. Databricks is exploring the idea that databases don’t have to be tightly bundled systems where storage and compute live together and your data is locked inside a proprietary engine. Instead, data can sit in open cloud object storage and the database layer can run separately and scale on demand. That makes operations lighter, you don’t have to overprovision for peak load, and things like cloning or scaling don’t require moving huge amounts of data. It also reduces vendor lock-in because the data isn’t trapped in a closed format, so switching tools or platforms later is less painful. https://www.databricks.com/blog/what-is-a-lakebase