Don't use an LLM for your README.md One of my more popular open source projects is Shell Bling Ubuntu, which, contrary to its name, actually supports many different operating systems these days: modern macOS, Alpine, Fedora 40 and up, Rocky Linux 8 and up, WSL (Windows Subsystem for Linux)...
RSS/Atom Feeds are Critical Robert Kingett tooted that indieweb folk are saying RSS readers are flawed, I'm here to say that feeds are ESSENTIAL to the web, all of it, and if any web gardener/creator/writer folk are indeed saying this, they need to re-evaluate.
OpenTelemetry traces: why deprecating span events is a terrible idea OpenTelemetry traces: why deprecating span events is a terrible idea
Qwen3.8 27B - Intelligence, Performance & Price Analysis Analysis of Alibaba's Qwen3.8 27B and comparison to other AI models across key metrics including quality, price, performance (tokens per second & time to first token), context window & more.
Stop reasoning blindly. How I tackle complex problems with benchmark driven development. - Charles AZAM Stop reasoning blindly. How I tackle complex problems with benchmark driven development.
Notes on purposeful skill maintenance/improvement/neglect Over the course of my life I’ve invested a lot of time and effort into the acquisition of different skills, abilities, competencies, specialized knowledge – for simplicity I will refer to all of it as “skills” in the following. Some of those skills, I wou...
Detecting goroutine leaks with synctest/pprof Explore different types of leaks and how to detect them in modern Go versions.
Harnesses are Situated Agents Harnesses manage the systems around agents, from the session, to the company, to the domain.
Session Visualization with Obsidian When working with coding agents such as Claude, Codex etc. we provide the prompts and context until we get to the final result. However a lot happens behind the scenes. The model constantly makes calls to tools, MCP servers, commands etc until it arrives ...
The Best Way to Make Your (Small) Vision Language Model Smarter Using test-time compute to boost the accuracy of small vision language models.
How I over-engineered my book | Ben Balter My book has a linter, ~5,500 automated checks, and a Pandoc pipeline that rebuilds five formats — EPUB, Kindle, paperback (×2), and web — on every git push. For a book one person wrote. Here's how I built and published it the way I ship software.
Lane Departure Microscope How a 7-inch digital microscope turned out to be running car dashcam firmware, complete with spoken lane-departure warnings and no speaker to play them.
Do That Which Makes Your Life Easy Why it might work well to make decisions (while building or maintaining software) in the interest of your quality of life?
The Artifact Is No Longer Proof of Competence The artifact still matters. It just can't carry the whole signal anymore.
The outstanding engineer who wanted to become a mediocre manager Why do so many outstanding software engineers become mediocre managers? A reflection on promotions, ambition, and why climbing the career ladder doesn't always mean moving up.
Harnessing the Power of Slop Driven Development — hank.bond When you're stuck on a design, letting an LLM implement the wrong approach can help you recognize the right one.
Attempting a game “Build and ship a computer game” is on many a programmer’s bucket list, mine included. However, I am quite cognizant of the reality that most people who claim …
you have arrived at your destination 😄 you should probably unsubscribe to this blog if you receive these as emails. love you fam. just hang out in person from now on. emails are dead now. k with that out of the way... https://docs.google.com/document/d/1-w...
AI;DR (AI; Didn’t Read) I'm about as pro-AI as you can be, but this is becoming a pet peeve of mine (and I'm not alone). That's why I love the AI;DR acronym as my new solution for ignoring the walls of slop.
TabBench-LLM TabBench-LLM evaluates large language models as few-shot, in-context tabular classifiers, head-to-head with two baselines.
Let the App Idea Die in the Chat One of my favorite things to do while I'm programming or working on my projects is to have slow, easy-listening jazz playing in the background. And ever since AI came along, I've been barely writing any code. I've probably built more small utility apps in...
Code is the Byproduct Recently, the Jacobian Conjecture was disproven by a counterexample discovered by an LLM. Shortly thereafter, a ChatGPT session from mathematician Terence Tao made the rounds online. In his chat, Tao uses ChatGPT to help him wrap his head around the impli...
A Preview of DuckDB v2.0 DuckDB v2.0 is coming this fall. In this post, we preview its headline features: DuckDB as a server, triggers, the VARIANT type, asynchronous I/O, a new SQL parser, a new storage format, and much more.
Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things Friday’s big release was Qwen 3.8 27B, an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba’s Qwen research lab. I’ve been looking forward to this one: 27B is an …
Who Are the Token Brokers? | Vectoral A look at the brokers buying unused AI credits from startups and reselling them — the marketplaces, the bulk-discount routers, and the message boards where off-market inference changes hands
Part 1/6 | Systemic Risks in the Managed PostgreSQL Industry: Extension Risks Are Real! Exploiting PostGis Memory Corruption Bug at NeonDB, SupaBase and Many More - Mehmet Ince @mdisec - Vulnerability Researcher | Building security products | Security Advisor | Amateur Muay Thai fighter Back in April, I was talking with our system and software engineering teams at PRODAFT about the possibilities of using a managed database service. Due to the nature of our business, we simply cannot start using managed services right away. I told my team...
Your harness design is probably bad. Harness engineering is the new buzzterm, but given its effectiveness, it's hard to deny the need of it. Most of the literature on the internet about harnesses revolve around coding, and sometimes proving math theorems, but often does anyone try to figure ...
Field Notes: Is Software Still the Point? The blog was quiet for a few weeks. I wasn't; at least not entirely. And of everything I read, built, and argued about over the summer, one idea kept pulling at me until I couldn't put it down. I want to think out loud about it here, before the new
Creating a Minimal Dark Factory | Joel Dare Field notes on shipping fast, privacy-first, evergreen web products.
Vim Fugitive in action - dzx.fr Learn how to harness the power of Vim Fugitive to efficiently stage, diff, commit, and resolve conflicts in Git repositories.
Why LLMs Are Unreliable Language Detectors LLMs can translate, explain grammar, and write fluently in dozens of languages — so it seems obvious they should be great at simply identifying what language...
Making stinkarm stink way less, or more? removing overengineered memory translation, hardening and new table driven armv7 instr decoding
Field Notes: oh-my-claudecode — The Orchestration Layer That Removes Humans from the Build Loop — IZHC oh-my-claudecode hit 19,754 stars this week. It's teams-first multi-agent orchestration for Claude Code — staged pipeline with plan, prd, exec, verify, fix. Socratic deep-interview before any code. tmux workers for real parallel execution. This removes th...
You can just choose how many bugs you want now There’s a bizarre aspect of AI coding that I’ve been trying to put my finger on, and I think it’s this: you can basically just decide how many bugs you want your software to have …
Double-double: 31 digits of precision without leaving the FPU If you ever need more precision than what 15 decimal digits of the double format can offer, and arbitrary-precision arithmetic is too slow, too heavy, or simply not available, there is a neat trick: glue two doubles together and treat them as one number. ...
Being ambitious and being a dad | Nicholas Charriere When I was going through YC, I didn’t mention my seven month old daughter to anyone. Now a few years later I have two kids, a dog and a very packed schedule. My kids are the best thing in my life. For years, my work was my life. Now my life competes with ...
Scaling RAG: Chunking, Reranking, and Cost Optimization — Darko Trpevski Ship RAG to production and watch it fail. Here's what works: semantic chunking, hybrid retrieval, reranking, and how to cut costs 5x.
Optimal Ask Let’s say that you are selling N widgets and you need to determine a price for your widgets. There are N customers, each of whom will buy at most one widget if your price is lower than the maximum price they are willing to pay. The maximum price that peop...
The Best Pattern is the One You Can Understand — hank.bond Incorporating LLM-recommended design patterns outside of your understanding should be avoided. Use only what you are able to comprehend.
A faster way to calculate the day-of-the-week A range of fast modulus techniques that beat compiler output
The Case Against Formal Verification, 50 Years Later - Ivan Gavran Writings on software correctness, AI, formal verification, and other technical topics.
EJ Labs — AI, systems, and infrastructure EJ Labs is a deep-tech consulting, AI product, and infrastructure systems studio.
How I stopped my Claude Code subagents from secretly running on Fable instead of Sonnet I run Claude Code with a big session model, Fable or Opus 5, plus a small zoo of subagents doing the boring parts. Gateway agents, formatters, checkers. The kind of mechanical stuff you pin to a small model once, in the agent’s frontmatter, and then never...
Writing code isn't the bottleneck anymore, reading is It doesn't matter how much code AI agents can generate. What matters is how much of it you can take accountability for. Good coding practices were never for the machines.
Why Pi Is My GOAT Agent Harness After moving from Claude Code to Codex and OpenCode, I found a minimal, extensible coding harness that I could finally shape around my own workflow.
Models Are Getting Dumber on Purpose - Walter van der Giessen Reasoning scores climb while per-token compute drops. Labs are stripping world knowledge out of models, and I think that's the right trade.
