5 Quick Links for Devs: Week 9, 2026
How I built Timeframe, our family e-paper dashboard
This is a great piece on tinkering with hardware and software to create something really useful for the family. While I'm not ready to build a dashboard for my family just yet, I loved seeing the thought that went into this.
Writing code is cheap now
For decades decisions have been made based on the expensive / time-intensive nature of coding. Simon W explores how our attitudes and approaches might change from here on out now that creating (most) code is cheap and quick. It's important to note that just because writing code is cheap it doesn't mean that the process of getting code into production is effortless - it just means that we can change where we place the emphasis.
Statement from Dario Amodei on our discussions with the Department of War
This isn't necessarily dev-related but I have to say I'm kind of tickled / terrified to be living in a time where this kind of statement is a real thing and not an unbelievable plot point in a middling sci-fi novel.
Related: Anthropic's Responsible Scaling Policy v3
Lessons from my overly-introspective, self-improving coding agent
Everyone's getting in on the action writing coding agents (that modify themselves!), and this is another great overview of the process and technical details that goes into creating a coding agent that runs on a loop, has a persistent memory and can respond to requests and proactively take action. The agents are often told to go off and create software, and the sloppification of the internet continues, but at some point genuinely useful personal agents will be available for every person on Earth, which is something to think about.
AI Code Review Gets Better When Models Debate: Claude vs Gemini vs Codex vs Qwen vs MiniMax
A common technique I've seen others use and I've used myself is to pass code and reviewed code through a multi-phase loop between different models - they tend to catch things that the previous models didn't catch, or change their output and opinions as the context they're working on becomes denser and more detailed. This post shows that having the models discuss code review improves the output quality.