Basil Berntsen

@basil404.bsky.social

Cloud architect by day. I like containers, openstack, terraform, ansible, AI IDEs, and black espresso. By night, I build Flowdoku Sudoku. I am a real person: handcrafted, small batch, organic posting only. No AI generated or automated communication.

I’m looking for testers for my sudoku game - it is thoughtfully designed to be the game I always wanted to play. if you’re interested, let me know and I’ll add you to the closed alpha test.

So I decided from the get-go that I wanted to make this game work out of the box on Android and iPhone. I ended up settling on building it in HTML5 and then wrapping it with Capacitor to produce native boundaries. it’s nice to have a single codebase!

Cool! I’m working on a big refactor of my code base, and I just discovered that GitHub copilot doesn’t “charge” me extra for long running queries. It’s 0.3% to use claude sonnet 4.6 even if it has to run for a long time. So now I use Claude code for small tasks and copilot for big ones.

I’m building pipelines to deploy to the app stores. This would be an absolute nightmare if it weren’t for AI. I’ve done work like this by hand and we’ve reached a significant tipping point... every time the runner finds some new way to fail, I just paste a run ID to Claude code and it applies a fix.

A super powerful technique using AI development, Claude code in my case, is after you have it write a bug report, systematically ask “is this an example of tech debt or code smell?”

programming bootcamp in 2026: “welcome to week 1. today we’ll learn how to open claude code” week 2: “prompt engineering fundamentals. please remember to say ‘please’ so the model doesn’t unionize.” week 3: “debugging. when the app crashes, we’ll paste the stack trace back to the ai and wait.”

I’m starting to learn which models are good at which tasks. Grok code fast (unlimited with GitHub copilot premium) is surprisingly good and fast with coding tasks, but it’s interestingly bad at documentation, putting far too much detail into architectural decision records for example…

I’m experimenting with doing two agentic agile stories at the same time on the same codebase. Sometimes they provide very naturally segregated commits where they can just be merged into main. Sometimes I spend more time untangling them than I spent doing the work.

So I want my sudoku game to be available on Android and Apple phones. Rather than trying to have two code bases, I decided to make it in HTML5 and simply wrap the game for each platform. Anyone have advice on that topic or experience to share?

The more I use AI tools for development the more impressed I am. I’ve learned so much with my side project I’m building in GitHub Copilot. I’m building a sudoku app with HTML, CSS, and native JavaScript- no framework.

TIL: when they say the flu vaccine is a bad match to the strains that are circulating this year, it’s still effective against severe infection and hospitalization. I just assumed it was a complete miss.

I’m looking for people who play #sudoku on their phones. I got frustrated with the apps available so I’m making my own, but I want to hear how other people play. Message me here if you’d be willing to have an informal chat about Sudoku :)

There’s a problem with the way #bluesky does #feeds. To me, the algorithm that generates topical content is part of the base service, not an extra. I agree that it should be open/customizable, but I think feeds should be redesigned such that running one doesn’t require a user to manage a server.

The purpose of IaC depends on the use case, but I claim that the primary capability of IaC is the generation of multiple distinct configurations, variants, from the same IaC source code. medium.com/@bgrant0607/... Do you agree? #infrastructureascode #gitops

Variant generation is the primary capability of Infrastructure as Code

Variant generation, i.e. the generation of multiple distinct configurations from the same IaC code, is the primary capability of IaC.

medium.com