If you only read one thingUgly Code and Dumb Things (8 minute read) âPursuing elegance in a vacuum leads to abandoned side projects or frameworks nobody uses. By contrast, clunky but functional code often comes with just the right compromises for quick iteration.â I found the iteration aspect interesting, as that one didnât occur to me yet. Ugly code makes you iterate faster. There is a time for ugly code, and there is a time for maintainable code. The article goes deeper and is well worth the read. https://lucumr.pocoo.org/2025/2/20/ugly-code/
Stories Iâve enjoyed this weekClaude Code Top Tips: Lessons from the First 20 Hours (19 minute read) Overall frame: âHereâs a mental model thatâs helped me: think of Claude Code as a very fast intern with perfect memory.â Very practical advice like âHow can you modify CLAUDE.md to prevent this issue in future?â Better than Cursor? Not sure yet, Iâll let you know. Different for sure. https://waleedk.medium.com/claude-code-top-tips-lessons-from-the-first-20-hours-246032b943b4 High Pay, Low Respect: 60% of Cybersecurity Pros Want to Change Jobs (57 minute read) Many insights into cybersecurity thinking, like the fight against âexecsâ that do not budget for cybersecurity, do not listen and seem to have cybersecurity professionals only as scapegoats when something happens. Be aware of this when you hire your first cybersecurity engineer (and remember, top skill: Getting other engineers to adhere to security, NOT: to have the very best security skills). â60% of cybersecurity pros are itching to jump shipâ keep that in mind, if you have them on staff, also: An opportunity. I know security is your weak area, a good article to get a deeper understanding in an easy way. https://www.vulnu.com/p/high-pay-low-respect-60-of-cybersecurity-pros-want-to-change-jobs Another framework for native UIs with web skills. âLynx delivers vibrant and engaging UIs for large-scale apps like TikTokâ I doubt this, Iâd think UI designers do deliver engaging UIs. That aside, itâs interesting they use Rust to make (all?) tooling, for a long time now Iâm arguing you should use the best language for your tooling, not the language you write the tools for. Is this going to replace React Native? Flutter? At least I think this is one to keep an eye on. And be prepared for that engineer who wants to rewrite everything with Lynx now. The Model is the Product (11 minute read) Last newsletter I had an article that argued, the model is irrelevant and the wrapper is the product, the model exchangeable. Here: âDeepResearch is not a standard LLM, nor a standard chatbot. Itâs a new form of research language model, explicitly designed to perform search tasks end to end.â Again, does that mean for you, you need to get into the model business - I would argue yes, you need to have the best model in your niche compared to your competitors. Stephanâs three AI steps: 1. Prompting 2. Fine tuning 3. Own model https://vintagedata.org/blog/posts/model-is-the-product A Plea for more Mikado (8 minute read) The Mikado method is one of the most important tools in your tool belt on technical debt. You need to master it. This article gives a good overview with practical examples. The key: "[R]evert your changes. Delete everything! And I really mean revert, not move to a new branch or squash. If you feel this change took you too long to just be deleted, it means it wasnât atomic enough and you need to split it." The Mikado method is about very small steps, â¤ď¸ it. Side note: This is also what refactoring is about - Refactoring is not those month long, change everything projects. https://dmathieu.com/en/opinions/mikado/ Railway V3: Faster and Cheaper (19 minute read) In the context of cloud providers, the author talks about the âhat on a hat business modelâ, e.g. running a cloud provider business on top of a cloud provider (there was once that game, Hats!) They now offer their cloud on top of their own data centers, 50% cheaper egress (this is where AWS gets you), 40% cheaper storage to their customers. Hat on a hat is a great way to start you business, at one point you too need to go âbare metalâ to stay competitive and relevant. Hehe, âand you sure as hell donât want yet another 3-letter acronym solving a made-up problem imagined by some marketing devfluencer.â https://blog.railway.com/p/launch-week-02-welcome Superintelligence Strategy (8 minute read) âWe introduce the concept of Mutual Assured AI Malfunction (MAIM): a deterrence regime resembling nuclear mutual assured destruction (MAD) where any stateâs aggressive bid for unilateral AI dominance is met with preventive sabotage by rivals.â Interesting. No opinion yet, but something I wanted to share. https://www.nationalsecurity.ai/ Git without a forge (21 minute read) Sometimes we use something and are so used to it, that we donât question it anymore - like Github. Git can be run without Github, heresy I know! The thing that would keep me on Github is their Auth and Github actions. For open source itâs discovery. But, we sometimes forget why we use something, and we might not need it. https://www.chiark.greenend.org.uk/~sgtatham/quasiblog/git-no-forge Delta Chat, decentralized secure messenger (1 minute read) Sometimes you have assumptions, that do not hold. Messaging was invented because it was âinstantâ - compared to email. Today many people might no longer want âinstantâ because of overload. And the internet got much faster (I started BBS with a 2400 modem and the internet on a 56k Zyxel, today we have Fiber into our apartment). What distinguishes Delta Chat is that it also works with emails as a transport layer. Your product might also make some outdated assumptions, and could innovate by dropping them. AI Code review is always wrong (3 minute read) If you think someone is stupid, think again, you might just misunderstand the person. Thinking someone is stupid is a great opportunity to learn, they might have a different context frame than you. Is AI code review âalways wrongâ? The author is unhappy but itâs hard to see if the AI is wrong, as the code presented is not the code reviewed. What I would be interested, what was the prompt (didnât say), and I have an urge the author misunderstood the review comment. But who knows? My experience with Claude in Cursor for reviewing my code (before committing) is great. Perhaps with AI, reviews happens before committing, not longer in CI. https://www.jessesquires.com/blog/2025/03/04/ai-code-review/ Join the CTO newsletter! | |
