If you only read one thingHow I Choose What to Work On (18 minute read) “I will only ever work on things I want to work on. I would rather be poor than make a lot of money doing something I hate” For a long time I wanted to make a career, and change things. That way I became CTO of several companies. When on track to become a COO/CEO at an eBay company, half way through a high potential future leadership program, I realized I would hate that job, it’s not for me. And I left and joined the startup of my wife. Since then I only work on things I want to work on, like this newsletter (which costs money to send out, and has no income ;-) so the article resonates a lot with me. Are you happy?
Stories I’ve enjoyed this weekThe Gruen Transfer is consuming the internet (5 minute read) The idea of the Gruen Transfer (or effect) is that shopping centers are built in a way that you forget why you got there. So instead of “I need a pair of shoes”, in, buy shoes, out, you forget about the shoes. You stay longer and buy more. This explains parts of the internet, e.g. Amazon throwing everything at you at once, to confuse you and stay. Or Instagram search, where they distract you from what you were searching for. Or Netflix of course. You felt it, now you know how it is called. Don’t use it! https://sebs.website/blog/the%20gruen-transfer-is-consuming-the-internet The 2025 AI Index Report (225 minute read) Report with many findings, some obvious, some not so, like “7. AI becomes more efficient, affordable and accessible” which shows that in 2022 you needed a 100B+ parameter model to achieve 60% in the Measuring Massive Multitask Language Understanding (MMLU) benchmark. In 2024 this fell to 4B with Phi-3-mini. Again it’s hard to predict hardware needs and energy in a field with rapid technical advances. Many other findings + some nice graphs. https://hai.stanford.edu/ai-index/2025-ai-index-report AI coding mandates are driving developers to the brink (31 minute read) Several surveys mixed results. Productivity goes up, developers' confidence in AI goes down. "[..] where every one of the company’s 40 engineers is now using AI coding tools in their work every day and reporting benefits. An internal survey revealed a roughly 40% productivity increase among the engineering team" - there seems to also have been a hard push against developers resistance (something I encourage CTOs to do too). Higher productivity means either building more or letting go of developers. Most companies I meet mostly build features with no impact, they struggle to come up with relevant features (for a variety of reasons, no vision being one of them, no understanding of their customers another one), so they will struggle even more to come up with relevant things. The gap between those companies that look at throughput will widen to those who look at impact, with AI the first ones will create more of those meaningless features, the second will have exponentially more impact over time. Which one are you? (Also, my theory of the productivity YoYo says, all developer productivity gains are filled with things developers have fun with but decrease productivity, like React and JS frontends in general. So counting on developers, those productivity gains will be short-lived) https://leaddev.com/culture/ai-coding-mandates-are-driving-developers-to-the-brink AI is going to hack Jira (20 minute read) Must read. Article got me with the sub-headline, mentioning “Agile Industrial Complex”, love it. And then delivers a punch with “In Big Agile, engineering = new features.” - yes! Reading my newsletter though, you know the article lost me with “And AI is about to make it all much worse.” AI will do away with software engineering (remember AI != software), so everything the article writes about engineering and infrastructure is irrelevant. Still, must read, even if wrong. https://blog.godfreyai.com/p/ai-is-going-to-hack-jira What the latest DORA report says about GenAI and Software Development (26 minute read) "[..] the report also finds that increased AI adoption correlates with developers spending less time on tasks they find valuable. [..] In essence, it suggests that GenAI might currently be better at accelerating preferred tasks than eliminating shallow and tedious work." https://makemeacto.substack.com/p/what-the-latest-dora-report-says OpenAI now also has a CLI coding agent like Claude Code. Claude Code right now is excellent for one-shotting applications and for prototypes and MVPs (I wouldn’t let it change production code, but of course your risk appetite might be different than mine). Will check Codex over the next days, to see how it differs form Claude Code, for all interested there will be an article on my AI newsletter https://www.tabulamag.com/ - I guess there is not a lot of difference, but we’ll check. https://github.com/openai/codex Microsoft researchers say they’ve developed a hyper-efficient AI model that can run on CPUs (14 minute read) Again, as Popper said, the future is difficult to predict ( when | because ) breakthroughs are happening. I maintain my position, that we have not seen the final architecture or hardware to run AI on. We had graphics cards lying around, and they worked for AI, so we made NVIDIA the richest company in the world. But as NVIDIA replaced Intel, an AI hardware company will replace NVIDIA. Predictions about power usage are premature. Fintech founder charged with fraud after ‘AI’ shopping app found to be powered by humans in the Philippines (14 minute read) Title says it all. I guess it is no longer “Fake it until you make it” - at least if you don’t want to go to jail. Which is a good thing. Another one. ‘Wi-Fi Keeps Going Down’: Donald Trump’s Return-to-Office Mandate Is Going Terribly (20 minute read) WFH vs. RTO is still a thing? In the end, when things have calmed down, the CFOs will decide. https://www.wired.com/story/federal-workers-rto-chaos/ Join the CTO newsletter! | |
