If you only read one thingWhy most product planning is bad and what to do about it (13 minute read) âOur solution: Problem Driven Development, a 4-day quarterly process focused on identifying problems (not solutions)â If there is one thing you need to fix in your product development process, is that people think about problems not solutions. What I see in reality is companies where everyone has an idea, not for a problem, but for a solution - and then product/tech needs to execute. Great managers and CEOs think about the greatest problems that need to be solved to be successful, not come up with solutions and then search for a problem. Most âsolutionsâ I have seen from outside product/tech have no impact in the end. https://blog.railway.com/p/product-planning-improvement
Stories Iâve enjoyed this weekSoftware is Eating the World .. NOW! (2 minute read) âWhen Marc Andreessen wrote âsoftware is eating the worldâ back in 2011 [..] the second act is just beginning because software is now eating itself.â Have been very happy the last weeks, Sonnet 4.5 with plan mode works really well for me. Plan, revise plan, revise plan, tell it to make sure the plan works with the code, make sure it works with external APIs (re-check API docs) and the result is really good - if your own code is really good. What took a developer half a day can now be done in 15 minutes. âWe are seeing the emergence of a new category: micro-apps.â I wrote several small tools, like a âblogâ Go tool to manage my blog (which runs in Hugo), or a tool to manage the CDN Iâm working with (Bunny CDN). Never would have done that 5 years ago. No brainer today. Do I need to share those tools? No, itâs enough that they work for me. âIn this world, software becomes disposable.â And âCredits to Oliver Wehrens, Stefan Schubert-Peters, Daniel SchmeiĂ and Sebastian Heide-Meyer zu Erpen.â what a bunch of great people. https://devpg.substack.com/p/software-is-eating-the-world-now What Makes 5% of AI Agents Actually Work in Production? (12 minute read) There was the news that 95% of AI projects fail. While everyone focuses on that, the more interesting question is: How did 5% succeed? What can we learn? âThey all have one thing in common: human-in-the-loop design. They position AI as an assistant, not an autonomous decision maker.â and many more good points like âMemory Isnât Just Storage, Itâs Architecturalâ This is the richest AI article Iâve read for months and many actionable insights. https://www.motivenotes.ai/p/what-makes-5-of-ai-agents-actually In the wake of the NPM phishing attack, which nearly led to a catastrophe, there is turmoil in Ruby Central about the Gem management infrastructure. Some people broke away with gem.coop. Some thoughts: First, is your supply chain safe? Second: Why Javascript? (just kidding, but seriously, why?) Third: Taking away from Open Source often ends in a disaster - or doesnât it? Redis seems fine, as does Terraform. Structured Procrastination (7 minute read) Iâm a master in that! âI am working on this essay as a way of not doing all of those things.â Are you too? Still this newsletter went out and I had some very successful blog posts written this week. https://structuredprocrastination.com/ NIRS fire destroys governmentâs cloud storage system, no backups available (5 minute read) âHowever, due to the systemâs large-capacity, low-performance storage structure, no external backups were maintained â meaning all data has been permanently lost.â I sound like a broken record, but if you have all backups in one location (physical, or AWS/S3) youâre doing it wrong. Recently a client told me about a shock moment when the Azure project was empty, backups and all. Language Agnostic Programming: Why you may still need code (6 minute read) âEven with all that, I think that one aspect of programming will remain essential: debugging. No matter how good AI gets at generating code and even at debugging it, weâll still need to understand what that code actually does when it doesnât work as expected. And for that, we need programming languages.â Disagree of course, for that, âwe need testsâ. https://joaquimrocha.com/2025/08/31/language-agnostic-programming-why-you-may-still-need-code/ Asked to do something illegal at work? Hereâs what these software engineers did (12 minute read) Iâd wish more engineers would take a stand, many wonât and just do illegal things when told so. Like the Volkswagen engineers during #DieselGate. Some examples of engineers who went along with the crime, but âThe Director of Engineering questioned whether creating and using such a data set was legal, but Javice tried to assure the engineer by claiming that this was perfectly acceptable in an investment situation [..] The Director of Engineering was not persuaded and told Javice and Amar that he would not perform the taskâ Iâd wish more engineers would take pride in their profession and refuse to do illegal things. https://blog.pragmaticengineer.com/asked-to-do-something-illegal-at-work/ daniel:// stenberg:// (@bagder@mastodon.social) (9 minute read) âJoshua Rogers sent us a massive list of potential issues in #curl that he found using his set of AI assisted tools [..] I have already landed 22(!) bugfixes thanks to thisâ As Iâve been saying for quite some time now, if you only use AI to generate code or requirements or tests, youâre doing it wrong. Claude Code is an excellent analysis tool. https://mastodon.social/@bagder/115241241075258997 This Is How the AI Bubble Will Pop (10 minute read) Will it? My dislike is with the word âAI bubbleâ, what does it mean? The usage of AI? AI companies? AI users? AI investors? When the internet bubble popped in 2000, it was companies like petsDOTcom who were too early and not frugal enough to survive, which was the bubble bursting. Internet usage grew and grew and grew - I used more of the internet and did more with the internet during the time the âbubble burstâ compared to before. Plus, can Amazon or Meta who spend these billions on AI, really go belly up? What will happen is investments by VCs go down, and many companies will pop because they have no business model - either because their product does not work, is too early or OpenAI just built an agent builder. But this sounds like petsDOTcom âThinking Machines, an AI startup helmed by former Open AI executive Mira Murati, just raised the largest seed round in history: $2 billion in funding at a $10 billion valuation. The company has not released a product and has refused to tell investors what theyâre even trying to build.â https://www.derekthompson.org/p/this-is-how-the-ai-bubble-will-pop I built ChatGPT with Minecraft redstone! (18 minute read) Plain amazing, someone built an LLM in Minecraft, with training, chat and everything. What a hacker in the true sense of the word. I applaud you! Nightmare to debug I guess. https://www.youtube.com/watch?v=VaeI9YgE1o8 Increasing your practice surface area (18 minute read) âThe difference between being good and being great isnât talent or formal training, but the invisible practice that happens when youâre just living life.â Why did I put this article in the newsletter? It helps you understand how you can and grow your direct reports to the point of them becoming you. The how is in that article, believe it or not. LLMs Are the Ultimate Demoware (4 minute read) âWhen developing demoware, which I have done in the past, you develop a basic prototype of software and find a particular scenario in which it performs quite well.â Fits my model of Idea â Prototype â MVP â PMF â Traction. âIn the demo, you can say âWith this new dashboard, weâll look at X salesperson. The dashboard shows theyâve been performing significantly above average, so maybe they deserve a raise!â This is the reason of the prototype - what the article would call demoware, itâs not an MVP. The idea is to have something to show around and gather feedback, build desire and momentum. âLLMs, on the other hand, can provide seemingly excellent results in a demo environment with minimal work.â You need an Idea â Prototype â MVP â PMF funnel in your company for new features and products. https://blog.charliemeyer.co/llms-are-the-ultimate-demoware/ What is Stephan doing?Optimized my website, made it clearer, added TL;DR to articles. Mostly using Claude Code - which wrote a tool for me called âblogâ. I run my website on Hugo, but wanted more tooling around one part of it, mu blog (round SEO/GEO, listing articles, managing drafts etc.). In the past I would not have written all those small tools (I wrote another one to manage my CDN, Bunny CDN), being anxious of maintenance after coming back some months later - I feel this is no longer a challenge with AI. Everyone seems to be unhappy, Iâm very happy. Perhaps itâs also because my main language is Go, which leads to binaries I can easily use, has the right amount of type system to make AI work, but also is simple enough for AIs to understand - with a wide body of example code to be trained on. So I might just be lucky with my choice of language. Website is better now (but far from done). Events for Engineering ManagersJoin the CTO newsletter! | |
