Weekly notes: Getting an agent to write better tests in python, and 1 startup's platform tooling choices and how those felt

• The AI Vampire: AI people are working harder and longer to keep up with outcomes described by ai influencers.
• Quamina + Claude, Case 1: Tim Bray talks about a recent experience with claude that was helpful setting aside the feeling that ai is causing a lot of harm within the bounds of this experiment.
• Tips for getting coding agents to write good Python tests: Getting an llm to write better python tests by getting a test suite started with a preferred style or by asking it to look at a github repository with good tests can work. Another test hack is “extract the common setup into a pytest fixture”
• 7 learnings from Anders Hejlsberg: The architect behind C# and TypeScript: “Shared outcomes over personal preferences” and “Get to fast feedback first” are my favourite take aways from this.
• So what’s the next word, then?: High level notes about how a transformer type neural network works. How to represent language and attention are interesting here. Amazing how well these simple ideas which strung together work in practice. “Look at everything at once”
• (Almost) Every infrastructure decision I endorse or regret after 4 years running infrastructure at a startup: A long list of software tools somebody used in a startup company they helped to build.