My dearest gentle reader, I recently got featured on the Django Chat podcast as a guest. I have been following the podcast for a while now, and this was so much fun to do and chat with friends.

The audio-only version can be found here, and you can watch it here:

OpenAI will acquire Astral: Astral is an open-source, permissively licensed company that builds high-performing tools for developers in the Python ecosystem, such as Ruff and UV. We will soon find out what that means for the ecosystem, but it does reintroduce the topic of open-source funding. Open source is the sauce that keeps giving, but maintainers are human with finite energy and sometimes need some financial backing to keep doing the good work.

The PERFECT code review: How to reduce cognitive load while improving quality: PERFECT is an acronym as illustrated above to transform code reviews from subjective, cognitively draining tasks into a structured, high-value process. The 7 key areas are that the code fulfils its purpose, handles edge cases, is reliable in terms of security and performance, adheres to the architecture form, provides evidence by passing tests and ensures clarity of intent while managing personal taste without blocking progress. The goal of the PERFECT review is to establish clear, documented conventions and automate routine checks that focus on business value whilst offering constructive feedback rather than personal preference, thereby increasing developer morale and improving software quality.

The Black Box Problem: Why AI-Generated Code Stops Being Maintainable: When teams use AI in their codebases, they often experience a boost in productivity, but then experience a decline in maintainability because the AI code lacks architectural structure and intent. This is what is known as the black-box problem. The AI models prioritise speed and immediate correctness within a single context window, meaning they focus on making a single snippet of code work right now, without considering how it fits into the rest of your system. Yonatan argues this isn’t about improving the AI models themselves, but about improving the environment in which they generate code. True maintainability comes from a structured approach that enforces clear component boundaries, uses typed interfaces, and integrates automated testing during the creation process.

14 More Lessons from 14 years at Google: Addy emphasises that great engineering is less about writing code and more about making good decisions. This includes what to build, what not to build, and how to work effectively with others.

What I learned from the book Software Engineering at Google: Milan
highlights that the book is less about writing code and more about building and maintaining healthy systems over time. The idea is that software engineering is programming integrated over time. Developers must think beyond immediate functionality to long-term evolution, scalability, and team impact. The main lesson is that strong engineering culture, clear processes, automation, and psychological safety matter as much as technical skill in keeping large codebases sustainable.

AI Slop vs Constrained UI: Why Most Generative Interfaces Fail: AI doesn’t understand intent. It resolves constraints, so when those constraints aren’t explicitly defined, the result is inconsistent and hard to maintain. Successful interfaces aren’t fully open-ended. They are constrained UIs that guide the model with clear rules, components, and states, enabling reliable, repeatable outputs instead of fragile, one-off generations.

Your job is to deliver code you have proven to work: Simon argues that in the age of AI, the responsibility of a developer hasn’t changed. The job has always been to deliver code that you have proven works; developers must treat proof of correctness as part of the work itself. This means using testing, reproducibility, and evidence to show that changes behave as expected. Far from removing accountability, AI reinforces that developers are still responsible for quality, correctness, and the integrity of the code they ship.

Nobody Gets Promoted for Simplicity: simplicity isn’t naturally rewarded. You have to actively communicate the decisions, trade-offs, and judgment behind them.

Fast Software: More Programmers, Not Fewer: Speed and accessibility will rise dramatically over time, but at the cost of traditional software craftsmanship, which will change what it means to be a software engineer. This shift increases demand, but for a different kind of role: engineers acting as AI operators who generate and customise software quickly.

AGENTS.md outperforms skills in our agent evals: Giving AI agents structured context is always more effective than relying on on-demand tools like skills. The agents.md file achieved a 100% success rate, while the skills failed because the agents did not choose to use them. AI agents work best when given a clear, persistent structure upfront, rather than being expected to dynamically figure out when to access the right knowledge.

The word of the day is immaterial. Immaterial refers to something that is either unimportant and irrelevant or something that lacks physical substance.

Example in a sentence:
Whether you arrive in a limo or on a bicycle is immaterial to the host; what truly matters is that you show up to celebrate.

I am making my grand departure into the unknown.

Take care of yourself!

Until the next fortnight, my treasured reader, go forth, and may the odds be ever in your favour! 👏 🤖 ✊ ☠️ 🏹 🪖

Do you enjoy the latest issues of my newsletter? Buy me a piping cup of hot chocolate today!

Cup of Hot Chocolate