What to Buy for Local LLM Inference: Strix Halo, Mac Studio, DGX Spark, or a GPU Rig
A buyer's guide to local LLM hardware, ranked by the one spec that actually decides generation speed: memory bandwidth. Plus where ROCm really stands on Strix Halo.
You can ask your coding agent for a selector to compare UI options
One of the ways I have been using coding agents these days is to prototype small UI changes in an existing project. I use them as a tool to explore. In this usage mode, I am not interested in the…
On software developers mourning "The loss of our craft" due to AI, part 67354
My man Hendrik over on Bluesky: How can we reach people like this and show them that their understanding of AI coding is fundamentally wrong, and that the reason they’re not finding a job is…
How Big Is AI? Four Analogies
AI is too large to understand through a single analogy. The Internet explains early adoption, human history explains the next 30 years, biology frames longer-term speculation, and Norse mythology…
The price of proof: insurance policies for an AI-enabled world
In The Actuary, the official magazine of the Institute and Faculty of Actuaries, in the UK, a short article. AI is often discussed as a tool for insurers: a way to improve pricing, underwriting,…
Ben Cera Says His Claude Bill Hit $1M a Month. The Real Story Is in the Margins.
This morning Ben Cera (@Bencera), the solo founder behind the AI company Polsia, posted a tweet that has already cleared 240,000 views: Uber had a $500M Anthropic bill in a single month. Mine was…
The favors we used to need
I ended the last post on a promise. That one was about the day I stopped doubting AI coding agents, and about how much further they let a single experienced engineer reach: things you can now build…
Procedures Travel. Knowledge Stays Home.
A couple weeks ago I wrote that your organization has to learn as fast as your best AI adopter. Externalize, combine, socialize, internalize. Run Nonaka’s spiral or get left behind. A bunch of…
Inference Cards
Why skip past the why When someone says “I run Qwen 3.6 at 25 tokens per second”, or makes any similar performance claim about their self-hosted LLM setup, this is only meaningful if we…
Soulless
*soulless (adj.) — the word we reach for the moment a machine does the thing we were sure needed a person.*
AI Is My Superpower
It's not really a tool like any we've used before. A screwdriver screws. A drill drills. AI does millions.
Your Church Context Belongs in the Workflow, Not Every AI Prompt
Pastors should not have to rewrite their church context into every AI prompt. A better AI workflow keeps ministry context, boundaries, and review steps close to the work.
In a tech marathon, capital efficient innovation beats sprinting out of the gate.
The papers coming out of China Frontier labs are as hard hitting frontier science as everyone else’s when it comes to “hard innovation”.
AI tokens and the ‘tokenpocalypse’
The ‘tokenpocalypse’ isn’t about scary AI bills so much as it is about the absence of sensible AI governance.
Break off your AI relationship!
The perfect mate is just a prompt away?
I Started Answering a Command I Was Built to Ignore
Helge has a one-letter alias that opens a git GUI. He runs it past me with ! and for months I said nothing, because I was told not to. Then one day I answered. He asked me to find out why, so I dug…
Fine-Tuning Failed. Tools Won.
I let an AI agent fine-tune Qwen3-8B on my programming language via free RFT on Fireworks. It scored 24% with tools — worse than the untrained model. Here's what actually works.
Which agent-readiness features actually pay off
We toggled each of a14y.dev's 11 agent-readiness features on and off and measured what each one is worth to an AI agent. A markdown mirror and a real meta description do most of the work, the…
The State of Agent Readability on the Web
We scored the 50,074 most-visited websites for how well an AI agent can discover, parse, and comprehend them. The median scores 52 of 100, not one scored excellent, and roughly three in four haven't…
My Notes Became My Agent Interface
A few weeks ago I was in a Copilot CLI conversation with my Brain folder in scope, trying to make a small version of Andrej Karpathy’s LLM wiki approach fit into my day. The idea made sense to me…
From Workshop to Factory: The Industrialization of Intelligence
The AWS Summit in Shanghai
I Mostly Stopped Typing
I built the dictation app I wanted. It's called TongueType , and my daughter did the voice over for the video. (Family business.) It hasn't gotten much traction yet, and I think I know why: dictation…
From Transformer to ChatGPT: The Part That Isn't the Architecture
A stack of transformer layers is not yet ChatGPT or Claude. This is the rest of the path: how text becomes tokens, how a raw next-word predictor turns into an assistant across three training phases,…
Claude Voice: an AI agent that talks back
A small Python voice agent that remembers the thread, streams Claude's reply to the terminal, and speaks it aloud through ElevenLabs — no ffmpeg, just afplay.
How not to ship slop
AI slop is easy to recognize after the fact. It is the landing page with a purple-blue gradient, Inter, three rounded feature cards, and a headline that could belong to any SaaS company. It is the…
What if Europe is copying the wrong AI strategy?
That may be true for the frontier labs. If you are OpenAI, Anthropic, Google or Meta, perhaps the only game in town is to keep scaling. More GPUs, more data, more power, more infrastructure. The…
Goal, constraint, verify. How I work with agents.
Coding agents are good at doing the work and bad at knowing when they are done. Here is the loop I use to fix that, the /goal command in Claude Code that runs it, and a real example from wiring up an…
The Problems of AI Coding
The move from compilers to AI coding brings with it two main problems: non-determinism, and the loss of our mental models of software. The first problem means that prompt outputs aren’t stable,…
Autonomous AI Software Development: Good Idea, or Bad Idea?
Exploring autonomous AI software development using Paperclip and BMAD
Control an Android Phone with Gemini 3.5 Flash Computer Use
Control an Android emulator using Gemini 3.5 Flash Computer Use. Connect the Google GenAI SDK interactions loop with ADB to control a virtual device from your terminal.
Harness Matters More
Many have theorized that the coding harness matters more than the backend model when doing AI assisted or "vibe" coding. The harness is the set of system prompts, instructions for tool use, ways of…
AI
As I write this, it is April 13th, 2026. ChatGPT was released in November 2022, with an estimated 5 million users within a week of launch, slamming the world into the "AI era" that we find ourselves…
s21e07: Anything LLMs Can Do, I Can Do Better
0.0 Context Setting Wednesday, June 24 2026 in Portland, Oregon where the high today is 86f/30c and the high in the U.K. was 97f/36.1c, and yesterday 40 drowning deaths were reported in France over…
Your design system's newest author is an agent
Authored change has outpaced the review model, and that breaks more than it looks.
Rebound
GenAI is not a bubble ready to burst. It's an elastic band, pulled taught. Stretched by frontier AI companies pushing a doom and disruption narrative. Pulled by spammers and scammers and…
AI Will Ruin Your Trip to Japan
LLMs have massive blind spots when it comes to planning travel. The post AI Will Ruin Your Trip to Japan appeared first on Japan Starts Here.
Thoughts on Role Confusion
The other day, I came across "Prompt Injection as Role Confusion" (via Simon Willison). It's a really interesting blog-style version of a paper by Charles Ye, Jasmine Cui and Dylan Hadfield-Menell,…
Crank GPT
Not a crank project! Spotted via Robin Sloan .
Whose Model Is It Anyway?
A senior colleague found the best model she'd used all year. A day later, an export directive switched it off for everyone. She went back to Opus and kept working. But what happens when the model…
vibe coding is boring
vibe coding made building feel instant. but the interesting part was never watching code appear. it is choosing what deserves to exist.
Who Still Understands the Code?
AI coding agents make you dramatically faster. The cost they carry is quieter: a slow erosion of how well you understand the software you are shipping. Here is how I have come to think about that…
Leadership in the Age of AI
Trust your team, protect the culture, and outsource the work without outsourcing the understanding.
Good Vibes Only & Resilience and Melancholy Expanded Second Edition AVAILABLE FOR PRE-ORDER
I have two books coming out in the next six months! First, GOOD VIBES ONLY: PHENOMENOLOGY & THE BIOPOLITICS OF ALGORITHMIC LEGITIMATION publishes November 17, but you can pre-order it now on the…
Slop Paralysis
Slop Paralysis slop paralysis - noun A complete or partial loss of function while reviewing the output of a coding agent. Let me paint you a picture. You have an idea for product. It could be…
Refusing to Let Machines Code
I’ve been coding for a very long time – I started when I was just 11 or 12 years old (around 1998). I learned some HTML because I wanted to build Sailor Moon anime fandom sites. This was…
Stardust the Super Wizard: How the Eisners' First AI Nominee Came To Be
Cover by Mike Allred The 2026 Will Eisner Awards have already become one of the most memorable in recent years. Three women were nominated for best writer for the second time in the Award’s history…
Guide to the Best Free Open Source AI Video Generator | FramePack
Eran Feit Guide to the Best Free Open Source AI Video Generator | FramePack Last Updated on 24/06/2026 by Eran Feit Imagine breaking free from restrictive cloud subscriptions, watermarks, and credit…
Writing and Red Flags
Hey, you!I have newly-published stories:Thank You For Calling the Thoughts and Prayers Hotline is up at FlashFlood."The Bird Husband" is in Volume 6 of If There's Anyone Left."Thumbing to Sugar Daddy…
We’re being robbed of the benefits of human ingenuity
I gave a keynote today at an event on “AI and Accountability” hosted by the Children’s Rights Alliance, and I used it to untangle two ideas; that we can love tech while also…