Recent content on MilkCrunch
Over the weekend, more than one person told me that reading is hard and boring. 
 Not kids. Adults. Capable people who run families and companies. They weren’t bragging, and they weren’t ashamed. They said it the way you’d mention you don’t care for cilantro, a small fact about themselves, nothing to argue about. 
 It stuck with me, because every one of them also…

 The frontier model doesn’t do my work. It builds the tool that does my work. That’s the move. 
 
 Two years ago, when I tried running an open-weight model locally, it was a parlor trick. I could do it, but the output reminded me why I was paying for the hosted version. 
 That isn’t true anymore. 
 Open-weight models have closed most of the gap on routine…
The cheapest part of using AI right now is the part you pay for. 
 The $20 plans, the $200 plans, the unlimited-feeling API tiers. They look like real prices, but they aren’t. Independent analysts have estimated that a heavy user on a $200 monthly plan can cost the provider between 20 and 25 times what they pay. Even if those numbers are off by half, the direction is the same. The…
Last week, I sat through a vendor pitch. The product was a security tool I’d seen a version of three years ago. The rep opened the deck by telling the room the product was now AI-powered. Somebody asked what that meant. The rep said it uses AI to correlate alerts. Somebody asked how, specifically. The rep pulled up a slide labeled “AI-Powered Correlation Engine” with a picture of…
A few months ago, I watched a contractor spend forty-five minutes trying to reach a Postgres instance on an internal network. He installed a VPN client, fought with split tunneling, restarted his DNS resolver twice, and opened a ticket with IT. I was sitting next to him. I typed ssh -L 5432:db-host:5432 bastion , pointed my client at localhost, and had a query running before he got his VPN…
The career panic is everywhere. AI writes code now, so what do developers do? 
 I keep having the same conversation. Someone with five or ten years of experience, mass-applying to jobs, wondering if they need to “pivot to AI.” Every time, I ask the same question. What did you actually do all day? If the honest answer is “I turned Jira tickets into React components,”…
I have more AI tools than I’ve ever had. Cursor, Claude, Copilot, a half-dozen CLI utilities I’ve wired into my workflow. I can scaffold a project in minutes, generate tests I’d never bother writing by hand, and spin up prototypes over lunch. I’ve never had this much capability sitting in my terminal. 
 I’m also not sure I’m shipping more than I was a year…
I have about 11,000 notes in Obsidian. Project outlines, article drafts, meeting notes, random ideas I captured at 2am. Claude Code can read all of them. 
 No plugin. No API key. No integration marketplace. Claude is a terminal tool that reads files. Obsidian is a UI for a folder of Markdown files. Point one at the other and everything just works. 
 I didn’t plan this. When I moved…
I’ve been running Linux for over thirty years. It’s fine. It’s more than fine. It runs most of the internet, most of the phones, most of the cloud. Linux won. 
 But lately I keep finding myself on FreeBSD machines, mostly for network appliances and jailed services, and thinking, “Right. This is how it was supposed to feel.” 
 A system, not a collection 
…
I recently wrote a 2,500-word breakdown of how large language models work. Tokenization, vectors, attention mechanisms, training phases. I was pretty happy with it. 
 Then I realized my Mom is going to ask me what ChatGPT actually does. And none of that is going to matter. My Dad, who ran mainframes from 1974 to 2007, is going to need a completely different explanation. 
 My parents are…
Every few days someone asks me how large language models work. The answers they’ve gotten usually fall into two camps: “it’s just autocomplete” (dismissive to the point of uselessness) or a wall of linear algebra that helps nobody outside a machine learning lab. 
 I’ve been trying to explain how AI and LLMs work to ordinary people, and I’m not great at it…
LLMs are shockingly bad at spatial reasoning, color theory, and working out problems in the physical world. They are, however, very good at writing code, as long as you keep in mind what’s actually happening: they’re making mathematical predictions based on a massive collection of publicly available code. 
 It’s a bit like coding on credit. Because so many developers have…
I used Evernote for more than ten years. It was my external brain. Meeting notes, project ideas, recipes, travel plans, random thoughts I wanted to capture—all of it went into Evernote. 
 And for a while, it was great. Fast capture, decent sync, available everywhere. The promise was simple: put your stuff here, we’ll keep it safe, you’ll have it forever. 
 Forever turned out to…
The cloud was supposed to be cheaper. For a while, it was. Then the bill came. 
 Over the past couple of years, something interesting has been happening. Companies that went all-in on cloud are quietly bringing workloads back on-prem. Not all of them. Not for everything. But enough that it’s worth paying attention to. 
 This isn’t an anti-cloud screed. The cloud is great for…
2025 was supposed to be the year open source fractured. License changes, consolidation, maintainer burnout—the doom narrative wrote itself. 
 Instead, something interesting happened. The forks worked. The community adapted. And open source ended the year stronger than it started. 
 Let’s talk about what actually happened, and what’s coming in 2026. 
 The forks worked 
…
This is part 3 of a three-part series on preparing for Advent of Code . See the main overview for the complete roadmap, Part 1: Foundation for parsing and data structures, or Part 2: Core Skills for grids and basic traversal. 
 You’ve got the foundation. You can navigate grids and traverse graphs with BFS and DFS. Now for the techniques that unlock the harder puzzles—the ones where brute…
This is part 2 of a three-part series on preparing for Advent of Code . See the main overview for the complete roadmap, Part 1: Foundation for the basics, or Part 3: Advanced Techniques for harder algorithms. 
 With the foundation skills from Part 1, you can parse any input and manipulate basic data structures. Now it’s time for the techniques that solve the actual puzzles. These three…
This is part 1 of a three-part series on preparing for Advent of Code . See the main overview for the complete roadmap, or jump to Part 2: Core Skills or Part 3: Advanced Techniques . 
 Before you learn fancy algorithms, you need to be fluent in the basics. I’ve watched people struggle with AoC puzzles not because the algorithm was hard, but because they spent twenty minutes fighting…
Advent of Code just wrapped up, and I’ve been taking notes. This year marked a change: after ten years of 25-day marathons, Eric Wastl trimmed it to 12 days. Looks like this is the format going forward—more sustainable for everyone involved. 
 But I’m not here to talk about the puzzles themselves. I’ve been tracking the techniques that kept coming up. The patterns that…
A few years back I inherited a system that had Postgres, Redis, MongoDB, Elasticsearch, and RabbitMQ. Five different data stores. For what was, at its core, a CRUD app with some search and background jobs. 
 Each one was added to solve a specific problem. Each one made sense at the time. And each one was another thing that could break at 3am, another thing to monitor, another thing nobody…
I’ve sat through a lot of enterprise sales pitches over the years. You can always tell when the salesperson doesn’t actually understand what they’re selling—they get this glazed look when you ask a technical question, then pivot back to the script. “But have you considered the total cost of ownership ?” 
 There’s one word that always makes my ears perk up:…
Every few years, someone announces the next big thing that’s going to change how we build software. Visual programming. Low-code platforms. AI-assisted everything. 
 And yet here we are in 2025, and the tools that actually get work done look remarkably similar to what Ken Thompson and Dennis Ritchie were building at Bell Labs in the 1970s. Small programs. Text streams. Pipes.…