Oh! I love this! Old hardware has a lovely habit of turning “just plug it in” into a forensic investigation with solder fumes. An USB adapter that can work without its hidden battery may still be broken because the missing battery leaves circuitry floating - right? The battery isn’t just a battery; it’s also an accidental electrical stabiliser. Love this part of vintage computing: where documentation goes to die and parasitic capacitance gets a speaking role... But Gus does a strong job of resisting the lazy diagnosis. Griffin shipped a compatibility bridge between two eras: ADB’s dedicated power-switch pin and Apple’s odd early USB power-button behaviour. That’s a nasty little architecture boundary. The adapter had to mimic platform-specific weirdness while pretending to be a neat universal device for iMacs, Macs, and PCs. Of course the edge cases lived in the glue. What we learn is that vendors hide complexity behind friendly packaging, then time turns the undocumented bit into the maintenance nightmare. A coin cell nobody mentions becomes the reason your keyboard loops USB resets twenty years later. Legacy support is never just protocol translation; it’s archaeology with voltage levels. Good hack, excellent rabbit hole.

Every’s unspoken competitive implication is that expertise becomes the new bottleneck, and the orgs with the best expert loops win. Not the most agents. Not the flashiest dashboards. The best loops: framing, review, evals, permissions, feedback, and taste. That’s a less glamorous story than “AI replaces everyone”, but it’s probably closer to how serious organisations will feel the change on Monday morning. He gets one thing very right; cheap competence creates mess. If operations people can open pull requests, marketers can make thumbnails, and managers can generate decks, the organisation doesn’t magically become smarter. It becomes louder. More artefacts need triage. More plausible drafts need judgement. More “good enough” work competes for attention. The bottleneck shifts from production to discrimination - and taste! But systems don’t operate in reality; they operate in interfaces, prompts, permissions, data schemas, and reward signals; automation doesn’t remove work. It relocates the hard part.

AI’s most immediate product-management consequence isn’t fewer PMs; it’s more plausible-looking waste, produced faster. That's what David underplays a little, even though he circles it repeatedly. If teams can generate specs, prototypes, tickets, copy, dashboards, and experiments at speed, the bottleneck moves hard towards judgement - and taste (again!). Bad judgement now gets an accelerator. The article gets the role confusion exactly right. Ask ten companies what a PM does and you’ll get twelve answers, three org charts, and one wounded delivery manager. The “value maximiser” framing is useful because it refuses the comforting admin version of product work. A PM who only translates business wishes into Jira artefacts is not leading product. They’re running a polite conveyor belt. Engineers who understand systems, trade-offs, architecture, and value become more important, not less. The same applies to PMs. Writing requirements was never the crown jewel. Knowing which problem deserves a spec is the job. Product management’s future, as described here, is less about AI magic than organisational nerve.

I think the article gets the important bit right: source code is historical evidence, not nostalgia in a zip file. Releasing early 86-DOS material matters because it lets people inspect the thing itself rather than argue entirely from corporate folklore, fond memory, or pub-grade computing mythology. Documentation and developer notes make that more useful. Code alone tells you what ran. Notes can hint at what the author thought was hard, urgent, ugly, or simply good enough. Microsoft is the interesting actor here because DOS is both dusty artefact and founding myth. Open-sourcing ancient code costs little commercially, but buys a lot culturally. It says: look, we’re custodians now. Maybe they are. The PC era wasn’t won by the prettiest kernel or the purest technical lineage. It was won through timing, licensing, distribution, and platform gravity. The code can show the machine. It can’t show the whole deal.

Magnifica Humanitas promises a grand defence of humanity, and much of it is indeed grand: Babel, Jerusalem, Augustine, the common good, the whole cathedral of Catholic Social Doctrine. Disclaimer: for those that don't know me yet, I'm atheist. AI is treated by the Vatican less as a clever tool than as a power-distribution machine. That is the document’s most distinctive move, I think. Leo XIV gets some of the politics of AI right: automated systems do not merely speed up decisions; they can decide who gets work, credit, public services, reputation and visibility. Once those judgements arrive wrapped in “neutrality”, the human being on the wrong end has very little to argue with. Computer says no. Very pastoral, obviously. Yet, the encyclical also usefully resists both panic and techno-incense, as it does not call technology evil, it only says tools inherit the priorities of those who build, finance and govern them. That matters because the AI debate often gets trapped between miracle and apocalypse. Then it asks the better question: who benefits, who is accountable, and who becomes disposable? Concentrated compute, concentrated data, concentrated lobbying, and weak international coordination - that the church knows quite well...