Why building new software is continuous learning, programming with agents, the fallacy of AI job exposure charts, Siri’s business model alignment, and Google’s expensive infrastructure
- Because we’re building something new, in a time when software is changing (background vocals: everything is changing), and no one has a clue what the heck is going on - the most important thing is to embrace that and as soon as possible, as often as possible, ship things on which we can get feedback on, in a way that gives us valuable feedback by CI, by production, by our teammates, by select users, by select customers, by all of our users. https://registerspill.thorstenball.com/p/building-software-is-learning
- As agents write more code, the future of programming may depend less on typing code and more on giving agents strong, machine-checkable rules about what must always be true. https://blog.janestreet.com/formal-methods-at-jane-street-index
- Many people would like to analyse which jobs, companies and industries are most exposed to AI, assign scores, build charts, and map that against the progress of LLMs; but this is mostly impossible to do with much confidence, because we do not know how jobs will change, how companies will adapt, what else will change around them, or how to measure “work” cleanly; it feels directionally right to say that repetitive clerical work is most exposed, but that only helps if the exceptions are smaller than the rule, and we do not know that; in 1995, saying the internet would destroy the value of physical distribution for media was directionally correct, but it meant completely different things for music, newspapers, television and film; so the question is whether AI exposure analysis tells us anything useful, or just turns a truism into a chart - if half the jobs you analyse are unaffected, and other huge pools of work are transformed elsewhere, then the aggregate may be right only in the sense that, on average, we’re all dead. https://www.ben-evans.com/benedictevans/2026/5/24/ai-job-exposure
- Apple is keeping Siri fully integrated and Apple-controlled, owning the UI, voice experience, personal context, and Private Cloud Compute, while using Google behind the scenes to help run its cloud model (AFM Cloud Pro) and Gemini outputs to strengthen on-device models, meaning Google contributes raw compute and intelligence, but Apple owns the product and final experience. https://9to5mac.com/2026/06/08/craig-federighi-details-apples-collaboration-with-google-for-siri-ai-in-ios-27/, https://stratechery.com/2026/the-iphones-last-stand/
- Google is raising money by selling some ownership in the company, likely because building AI infrastructure is extremely expensive and it wants extra funding without taking on too much fixed debt risk. https://stratechery.com/2026/the-google-capital-company/ https://www.ft.com/content/2fe1fff9-f619-4a25-9b30-05bdc3bdea70