Week 16 · 2025-04-14 → 2025-04-20 · 11 newsletters
Patterns, Packages, and Pauses
ai-as-normal-tech · pattern-recognition-and-its-traps · supply-chain-and-systems · trade-war-and-the-cfpb
The week of April 14-20 was a quiet one in the inbox: fifteen emails across seven days, no dominant news story, and a heavy lean toward solo essayists rather than news outlets. The signal that did come through grouped neatly into four threads: a long, serious essay on how to think about AI, a small chorus of writers warning that our mental shortcuts are starting to backfire, two pieces on software supply chain fragility, and the first real assessments of where the new tariff regime and CFPB layoffs are heading. Not a week to overread.
AI as Normal Technology: The Week's Serious Essay
The week's heaviest piece by a wide margin was Arvind Narayanan's AI as Normal Technology, a 15,000-word paper arguing that AI should be understood the way we understand electricity or the internet: transformative but not categorically separate from the rest of technology, and definitely not a separate species we are losing control of. Narayanan's frame is three things at once: a description of where AI is, a prediction about where it goes, and a prescription for how to treat it. He rejects technological determinism and pulls forward the lessons of past technology adoption (slow, uneven, mediated by institutions) rather than the science-fiction framing that has dominated the discourse. The companion link to his earlier piece on why predictions of AI doom are overheated does the work of grounding the argument.
Two lighter pieces sat alongside it. Liz Tran at Life Skills with Liz Tran ran "Be Better than AI," arguing that since LLMs are pattern-matching machines that can only generate what they have already seen, the human edge is creating the pattern in the first place. Her line: "AI may be the best worker, but it will never be the genius." Gabby Lord at OMGLORD ran the honest "Before you claw your AIs out," opening with the admission that she is a late, begrudging adopter still trying to understand the tools.
The take: Narayanan's essay is the one that matters this week, and probably this quarter. The "normal technology" frame is a useful counterweight to the doom-and-utopia bookends that dominate AI discourse, and the operator implication is real: if AI is normal technology, then the right strategy is institutional adoption work, not existential risk mitigation. Tran and Lord are doing the personal-essay version of the same instinct.
Pattern Recognition and Its Traps: Jack Butcher's Trilogy
Jack Butcher at Visualize Value ran three pieces this week, which is unusual cadence even for him, and together they form a single argument about cognition. Traction Creates Distraction made the systems-thinking case that as your reach grows, your "surface area of attention" enlarges faster than your bandwidth can absorb, and you end up managing resonance instead of creating original signal. Pattern = Blindfold ran the shortest and sharpest: we survive by naming shapes, but the moment a thing fits a label its edges fade, and what is useful for survival is fatal for discovery. Fail Your Way Out of the Midcurve closed the trilogy with the practical inverse: theory is drag, overthinking is blindness disguised as a higher vantage point, and every misstep illuminates the edges of a path.
Read together, Butcher is making the same argument Narayanan is making in a different register. Both are pushing back on the cognitive default of fitting new things into old shapes. Narayanan's target is the "AI is a separate species" reflex; Butcher's target is the "I can think my way through this" reflex. Both end up in the same place: trust the slow, normal, hands-on work over the dramatic frame.
The take: this is the week's quiet through-line. If you only catch one piece in the trilogy, Pattern = Blindfold is the one to save, because it is the shortest and the closest to the actual mechanism. The combination with Narayanan is the real argument: stop letting your priors do the work your attention should do.
Software Supply Chain and Systems: Two Builds, One Attack
Jacob Voytko at Client Server ran two pieces with the most operational substance of the week. The first, on Bazel's tenth anniversary, made the honest argument that Bazel is for three groups (Blaze-pilled ex-Googlers, companies large enough to staff a build infrastructure team, and companies with insanely slow builds) and not for anyone else. The diagnosis is that Bazel was born from Google's separate evolutionary branch of engineering and "makes you do things the Bazel way," which is why the world has drifted away from its happy place. The references to Google Nomulus and Kubernetes ground the institutional point.
The second Voytko piece was the week's most actionable security story: Slopsquatting targets LLM coders with supply-chain attacks. Researchers found that 5-20% of package names hallucinated by LLMs are stable and frequent, which means attackers can squat on the fake names and wait. The proof-of-concept was a fake huggingface-cli package that received more than 15,000 authentic downloads in three months and was being recommended in repositories belonging to large companies, including one tied to Alibaba research. This is the inverse of typosquatting and a real, present-day exploit, not a theoretical one.
Aditya Bhargava at Ducktyped ran the soft-edge version with An illustrated guide to route tables, the next installment in his AWS networking series. It is teaching writing, not news, and exactly the kind of patient infrastructure explainer that holds up because the underlying material does not change.
The take: slopsquatting is the news item in this cluster, and it is one of those rare attacks where the right defensive posture (pin and verify your package dependencies; do not blindly accept LLM-generated install instructions) is the same posture you should already have. Voytko's Bazel retrospective is the right kind of long-term institutional read.
Trade War and the CFPB: The Week's News Items
Two pieces carried the political-economy load. The Last Bear Standing ran Into the Trenches on the state of the trade war nine days after the April 2nd Liberation Day announcement. The argument: the shock-and-awe opening worked better than critics admit, the 90-day pause was a real tactical retreat, and the administration has implemented the highest and most wide-sweeping tariffs in a century without immediate retribution. But the next move is impossible: negotiating a hundred bilateral trade deals in three months while fighting every counterparty in the same public forum, with the China faceoff turning into a bare-knuckle brawl. Trump has not held the steely composure he wore in the Rose Garden, and the stock and Treasury volatility has already cracked the posture.
Zarik Khan at Fintech Compliance Chronicles ran two pieces, with CFPB in Crisis: Mass Layoffs Reshape the Bureau the more consequential of the two. Acting Director Russell Vought announced mass layoffs at the CFPB on April 17, following an appeals court ruling that lifted parts of the previous freeze on reductions in force. The detailed breakdown of which of Judge Amy Berman Jackson's eight provisions survived and which did not is the rare piece of compliance writing that lives up to the genre. Khan's podcast with Fernando Castellanos of EverC from Fintech Meetup 2025 is the lighter companion.
The take: these two stories are the only macro items in the week, and they both point the same direction. The administration is using the executive bandwidth aggressively but losing the composure required to actually convert pressure into deals or institutional reshaping. The CFPB story is the more under-covered of the two and the one to watch, because the layoffs are the precedent for the broader civil-service campaign that is clearly coming.
Lifestyle Grace Notes
A few smaller pieces rounded out the week. Ben Kassoy at A Strawberry Spinning Like a Dreidel announced a writing workshop called Transformation Station for April 22. Lingthusiasm ran an episode on gesture, pegged to Lauren Gawne's new Slim Guide from Oxford University Press, including a memorable digression about a "tricked-out coloured garden relax chair" researchers built to make people say "um" more without tipping them off about the study. Internal Tech Emails surfaced Mark Zuckerberg's October 2013 internal note about Snapchat declining Facebook's $3.1 billion offer plus $1 billion in retention packages, with the additional detail that someone on Snap's board was telling Evan Spiegel that Snapchat could soon be worth as much as Instagram (which Zuckerberg pegged at $6 billion at the time).
Three Takeaways from the Week
The week's real argument was about cognition, not technology. Narayanan's "normal technology" frame and Butcher's pattern-recognition trilogy are arguing the same thing in different registers: the dramatic frames we keep reaching for are themselves the failure mode. If you take one operating principle into the next week, it is to treat the new thing as ordinary until forced to do otherwise.
The week's only live present-day exploit was slopsquatting, and it is the kind of supply-chain attack that does not require a fancy defense, only the discipline you already should have. The CFPB story is the political-economy item to track, and the trade-war analysis from The Last Bear Standing is the most clear-eyed read on why the next ninety days are harder than the last nine.
If you only revisit three pieces from the week, I would suggest Arvind Narayanan's AI as Normal Technology for the cleanest long-form frame on AI, Jacob Voytko's Slopsquatting piece for the most actionable security read, and The Last Bear Standing's Into the Trenches for the soberest read on where the trade war goes from here. A sparse inbox, but the pieces that landed earned their slots.