ICE and CBP overreach (ahem) are more prominent now, but Garret M. Graff demonstrates the problems spark from the creation of the Department of Homeland Security almost 25 years ago. “Since 9/11, the culture of ICE and CBP has meant that the agencies have been what you might call a fascist-secret-police-in-waiting, troubled agencies simply waiting for an ambitious would-be authoritarian.” Graff brings the numbers to support that statement, part of his larger review Accountability for ICE and CBP. Example: “Indeed, for much of the 2010s and likely before and since, it appears the crime rate of CBP agents and offices was higher PER CAPITA than the crime rate of undocumented immigrants in the United States.”
In Backseat Software, Mike Swanson shows how we moved from quiet software to the attention-hungry services which fill our notifications today. “Software didn’t break all at once. It eroded slowly, one reasonable justification at a time.”
Nobody knows how large software products work by Sean Goedecke underlines why even deterministic software is hard to understand. “The only reliable way to answer many of these questions is to look at the codebase. I think that’s actually the structural cause of why engineers have institutional power at large tech companies. Of course, engineers are the ones who write software, but it’s almost more important that they’re the ones who can answer questions about software.” AI coding tools help distribute this capability throughout an organization, but remain dependent on the documentation that’s barnacled in and about the decisions why the code exists in its current state. It’s useful and trivial to include semi-automatic documentation as part of AI workflows going forward. But you need to check your ground truth first!
I’m glad the Mayor of San Francisco is still recording his schedule and responding to FOIA requests. Public, elected officials are accountable when we hold them to these standards. It can’t be comfortable to share everything, of course, even when you’re proud of your work. Good on the SF Chronicle for asking for the data and then analyzing it: Here are the influential people S.F. Mayor Daniel Lurie spent time with during his first year in office.
Yes, if… from Michael Heap makes the tradeoffs explicit in choosing what work we do. Usefully, the phrasing avoid the conversation-ending, hackles-raising No in favor of just enough nuance to guide a conversation. I often say “straightforward, not easy” to force an acknowledgment that “easy” is not a useful judgment for planning work.
Daniel Hon winds around the topic of OpenClaw (by all its names). “People will do the stupidest thing in the easiest way to get the most utility out of something until it hurts too much, and then they’ll only pull back just a little bit.” That’s what grabbed me from the longish essay That Whole Moltbot/OpenClaw YOLOing AI Agents Thing.
I’m not a software engineer, but for a side project working with Claude Code, choices about storage are about to be relevant. It’s 2026, Just Use Postgres is from a vendor that’s built around Postgres, so bias is clear. But the simplicity appeals, since I’m not working on solving internet-scale problems with personal projects.
When I hit my Claude Code limits, typically a good reminder to step away from the keyboard for a bit. But Claude Code: connect to a local model when your quota runs out is another approach who are willing to set up some more local infrastructure and accommodate differences in model quality. Wait six months (three?), and this pattern may be more common.
Easy for a Gen Xer to agree with this one: We (as a society) peaked in the 90s.
OK, this one I only skimmed but if the link survives, maybe I’ll return. Sleeping In: A Short History on Sleep before the Industrial Revolution is from February 2024, and I’ve liberated it from browser bookmarks to this post. Sleeping straight through the night is a modern idea, or that’s my takeaway about broken sleep from the skim.
How is data stored? is a chapter in Dan Hollick’s Making Software, which is a lovely digital book that may get a premium printing in the future.
“If you can’t explain your idea clearly in writing, you probably can’t explain it clearly at all, and if you can’t explain it clearly at all, you probably don’t understand it well enough to build it.” The pitch deck is dead. Write a pitch.md instead. JA Westernberg uses AI jargon of the moment to make a timeless point. What is more timely is that now you really can start building something with less understanding because you can waste only your own time and tokens, first.