If you only read one thingWe Replaced Our React Frontend with Go and WebAssembly (24 minute read) Is WASM the future? This post is very convincing. You write frontend and backend code in the same language - which for whatever reason is not JavaScript or TypeScript for you - an then compile the frontend code to web assembly (WASM). In this case they use Go. But WASM has been around for a while and still hasnât caught on. Is is because of React momentum? Is it because of Javascript developer momentum? Is it because WASM is a bad idea? I wonder. As a CTO as this point, I would go to an intern and let them write a WASM prototype for our application to see if itâs a good idea or not. Good tip for CTOs: Write a small version of your app to try to see if the new technology is better than what you currently use. Do not willy-nilly change technologies. Also have a process around this, so it doesnât need to be debated every time you want to try something new. https://dagger.io/blog/replaced-react-with-go
Stories Iâve enjoyed this week(Almost) Every infrastructure decision I endorse or regret after 4 years running infrastructure at a startup (21 minute read) I appreciated posts like this, a long list of infrastructure decisions and the pros and cons of each. What worked, what didnât work. Usually you get one technology per blog post. Here we have dozens. Very useful if you start a startup or scale a company. The End of Programming as We Know It (27 minute read) Tim is a great guy. Iâve learned many technologies from OâReilly, Linux, Python, Java and many more. In the 90s they were the only game in town. This said I disagree with everything Tim writes in this article. It article starts of with âThereâs a lot of chatter in the media that software developers will soon lose their jobs to AI. I donât buy it.â Okay! Letâs read on. The argument seems to be that coding became easier over the last decades, from machine programming to Python, with libraries, frameworks and SaaS services. Still we see more and more developers. AI is just another tool. Then the argument breaks down. It takes factory workers as an example and âWorkers needed new skills not only to use the machines but to repair them, to improve themâ But if you look into a factory today, in most there are no people! Recently watched a video about a factory of a well known German tools brand. The person interviewed said when he started the floor was full of people. Today there are only a handful running everything. And people who moved out of factories moved into (services and) knowledge jobs. Where do people move to when knowledge jobs disappear? âLessons from history tell us that when automation makes it cheaper and easier to deliver products that people want or need, increases in demand often lead to increases in employmentâ Consumption today, and for some time, has been limited by the time you can spend to consume, not the money to buy the consumption. The biggest obstacle to new computer games is not the cost to produce them, but that people spend their time on TikTok Fortnite (âbiggest competitor to any new video game is Fortniteâ, Matt Piscatella). People do not use more websites or software just because it gets cheaper to build them. How many apps do you use on your phone? Software was a great tool to solve problems for 50+ years. AI will replace software (in the sense of something written with source code and code executed) as the best tool for any job. This means we no longer need the people to build those tools. Like smiths who smite swords. I stay with my prediction: AI will kill all software engineering. https://www.oreilly.com/radar/the-end-of-programming-as-we-know-it/ Revenge of the GPT Wrappers: Defensibility in a world of commoditized AI models (32 minute read) Hey! I also thought the defensibility was in the AI model, not the wrapper (the AI app of your startup). It looks today - and if you use for example Cursor this is clear - that AI models are just and exchangeable backend option. Gemini? Claude? OpenAI? Just a dropdown in a configuration dialog. This does not mean, if youâre going deep, you should not fine tune a model to your use case. But, hey, I was surprised by reading this and realizing how defensibility has changed. https://andrewchen.substack.com/p/revenge-of-the-gpt-wrappers-defensibility The Duolingo Handbook (13 minute read) Not a guide to become a unicorn, but some nuggets in there,âTake the Long View: If it helps in the short-term but hurts Duolingo in the long-term, itâs not right.â And perhaps the next time the CEO wants to take a shortcut to release earlier point them to âRaise the Bar: To change how the world learns, we must do world-class work.â https://blog.duolingo.com/handbook/ Deepseek vs Claude PR Reviews (8 minute read) This compares Claude Sonnet to DeepSeek R1 reviewing code pull requests. DeepSeek is better at finding critical bugs (+13.7%), is better at finding bugs in general (3.7x), has a better understanding of code and produces less noise. The time is near where all review will be done by AI and bugs going to be fixed. One will need to train the bot on your engineering culture though. Do you? https://www.entelligence.ai/post/deepseek_eval.html Youâre missing your near misses (23 minute read) âmost of your incidents arenât going to repeat incidents. [âŚ] The real challenge is preventing and quickly mitigating novel future incidentsâ Great article on what people miss (pun intended). Two things to not miss your near misses: 1. Whenever something looks strange in ops, you need to understand it. Donât shrug it off with âwell the website worksâ. And this might take time to change the culture, too often when I was an engineering manager looking into ops data, seeing strange things, people just shrug. Itâs like that IT crowd sketch where the IT guy looks at the desktop of his boss, shouts out âthe lady should not be thereâ and gets the reply âsheâs only there sometimesâ. Thatâs how I feel when I point at load spikes and someone says âyeah they happen sometimesâ. You need to understand what is going on. 2. Every post mortem digs deep. It starts with what what went wrong (tech level), what mistake an engineer made (blameless!) on a people level, what processes led to that, what culture led to that, and so on. On every level you find several problems. And you fix all of them. An incident is always the overlap of two problems. The database stops PLUS the failover does not work. The database loses data PLUS the backup does not work. All the things you fix on every level (not just the âoneâ root cause) are near misses. https://surfingcomplexity.blog/2025/02/01/youre-missing-your-near-misses/ Feature Flags vs Configuration Options â Same Difference? (31 minute read) Deep dive into configuration options. This is a research article and I fear our understanding of feature flags is way behind research. Go read it if you use or want to use feature flags - Remove feature flags, document feature flags, Testing, all included. Today my understanding of feature flags is: They decouple deployment from feature availability. This way you can make features available to users at the time of your choosing - for example in a grouping that makes sense and can be promoted. But you still can push the code to production at any time, no need for coordination. https://www.cs.cmu.edu/~ckaestne/featureflags/ What do I mean by some software devs are ângmiâ? (13 minute read) âware engineers who havenât adopted or started exploring software assistants, are frankly not gonna make itâ Very interesting idea in the article. Bad coders might use AI to be more productive and get better results, because âgoodâ developers think they donât need AI because they are good. This makes âbadâ coders climb to the top of performance reviews. âGoodâ coders will not get promoted, donât get salary increases, and will be let go. Didnât think of that. Wonderful new dystopic idea. Will this speed up or slow down AI adoption? DOOM running on Apple Lightning to HDMI dongle (21 minute read) Amazing. Also amazing, it looks like your Lightning to HDMI dongle contains a CPU and runs a stripped down version of iOS. In the dongle. A video https://www.youtube.com/watch?v=4XCkeN0XuqA - the author has been asked to port Linux to the dongle. Thatâs what brought us into technology. Oh the wonders. https://www.heise.de/en/news/Thanks-to-Mini-iOS-Doom-on-an-Apple-HDMI-Lightning-dongle-10277929.html Join the CTO newsletter! | |
