Today’s Recommendations: Developer Gripes, Vampire Science, and Europe’s Payment Revolution

Five more recent links from my bookmarks.

Claude Code Is Being Dumbed Down — Symmetry Breaking

I’m sharing this one because it’s a legitimate gripe worth paying attention to, even if it stings a bit given which tool it’s about. Version 2.1.20 of Claude Code shipped a change that replaced every file read and every search pattern with a single summary line, “Read 3 files.” Which files? Doesn’t matter. “Searched for 1 pattern.” What pattern? Who cares. The author’s frustration is well-founded: users asked for a simple toggle to restore inline file paths, and instead got a series of patches to verbose mode that nobody asked for. The broader point, that simplification aimed at the “average” user can make a power tool genuinely worse for the people using it most, is worth keeping in mind whenever we’re designing anything.

Why Vampires Live Forever — Machiel Reyneke

The best piece of writing I’ve read this month, which is saying something given the competition. It starts as a science essay about heterochronic parabiosis, the research showing that old mice connected to young mice biologically rejuvenate, and quickly becomes a mock-serious investigation into whether Peter Thiel and Bryan Johnson are actually vampires running a coordinated disclosure program. Recent research from UC Berkeley suggests that the benefit of young blood might not come from something in the young blood at all. It might come from diluting the old blood. The young blood doesn’t add youth, it removes age. Rigorously footnoted, genuinely funny, and it’ll make you look at the longevity influencer space differently.

Europe’s $24 Trillion Breakup With Visa and Mastercard Has Begun — European Business Magazine

A detailed look at something that’s been quietly building and finally reached a milestone. ECB President Christine Lagarde has warned that virtually all European card and mobile payments currently run through non-European infrastructure controlled by Visa, Mastercard, PayPal or Alipay, and that every card transaction sends European consumer data to the United States. The answer the industry is coalescing around is Wero, a digital wallet now backed by a coalition of 16 banks. Wero already has over 47 million registered users in Belgium, France and Germany, has processed over €7.5 billion in transfers, and a recent agreement with the EuroPA Alliance instantly connects approximately 130 million users across 13 countries. Relevant context for anyone thinking about payments infrastructure in Europe right now.

Magic Words — Dave Rupert

A short, honest reflection on the current state of AI “skills”, the text files that get optionally injected into context windows to guide agent behavior. Skills remind me of magic numbers in programming. Magic numbers are a pattern you typically try to avoid. They’re a code smell that you haven’t actually solved the problem, but found a workaround that only works in a particular context. And so now we have these “magic words” in our codebases. Spells, essentially. Spells that work sometimes. Spells that we cast with no practical way to measure their effectiveness. The unstable, non-deterministic nature of this bothers him as a craftsman, and I relate. The “hope-based development” framing has stuck with me.

How to Write a Good Spec for AI Agents — Addy Osmani (Elevate)

The most practically useful long-form post I’ve bookmarked this quarter. If you’re using Claude Code, Cursor, or any coding agent in a serious way, this is required reading. The key is to write smart specs: documents that guide the agent clearly, stay within practical context sizes, and evolve with the project. The five-principle framework is solid, but what I found most valuable is the three-tier boundaries model: distinguish between “always do” actions the agent can take without asking, “ask first” actions that require human approval, and “never do” hard stops, like never committing secrets or editing vendor directories. The section on breaking tasks into modular prompts rather than one giant context dump is equally worth your time.