- Microsoft’s core capability > building solid infrastructure and driving software efficiency. Every major tech shift - mobile, cloud, and now AI - builds on the same cloud-scale foundation of servers, networks, and data centres, so Azure’s infrastructure keeps earning as each wave rolls in. Microsoft then layers increasingly sophisticated software that squeezes more performance out of that fixed hardware, through model tuning, data-flow tweaks, and smarter scheduling, driving unit costs down for both the company and every customer sharing those machines. https://stratechery.com/2025/microsoft-earnings-microsofts-core-capability-amazon-earnings/
- Find the people that you think are interesting and ask what they’re working on; once you’re around them, get more ambitious, have ambitious plans, and when the inevitable dismissals come, you have to be immune to rejection, because ambitious ideas seem wrong before they seem obvious https://foundersatwork.posthaven.com/find-your-people
- We write and ship software in pretty absurd ways: a simple “open my garage door” app can swell to millions of lines of code, and tiny programs often drag in more than a thousand outside code packages from who-knows-where. All that extra mystery code makes security a nightmare. Many developers realise this can’t go on, but plenty have never known anything different, and even those who do rarely get the time to build things the right way https://spectrum.ieee.org/lean-software-development
- AI needs an implicit feedback loop. Just like how Facebook has insights from every like, tag, and click and how TikTok converts a single swipe into pinpoint preference data - no forms, frictions - just pure behavioural exhaust fueling the algorithm. That same approach is super helpful for modern AI. A wrong movie suggestion on Netflix might slip by, but one hallucinated answer from a conversational model breaks the trust instantly. The fix isn’t another survey - it’s wiring learning directly into each micro-interaction https://writing.nikunjk.com/p/the-missing-loop-in-ai
- When you identify an uncomfortable emotion, make a list of every tactic you use to dodge it, and then notice how each tactic ends up producing the emotion you were trying to escape - say, “I don’t want to feel like a failure, so I play it safe, which makes me feel like a failure.” This loop explains why our choices are often driven more by hidden feelings than by complex data, and why true clarity comes from welcoming the emotion instead of running from it https://x.com/FU_joehudson/status/1917987459809648668
What I’ve been reading (W21, 2025)
Microsoft’s capability, understanding emotions, finding interesting people, software’s bloat and missing loops in AI