On AI Text Watermarking
Recently, a new regulation of the European Union went into force that requires AI providers to watermark anything their models generate. While this has been common for image or video generation, text…
Small Language Models (350M Parameters)
I have had good results with small language models (such as 1.2B parameters) for interacting with my MCP server. I decided to try to see how small I could go—if 1.2B worked, would 350M work,…
The Elephant in the Brownfield
In recent weeks, I met several talented, driven, experienced engineers and leaders working hard to bootstrap their ideas into startups. I call them brownfield refugees — people who left or were…
[ Video ] How To Choose an Open Weights Model – Episode 3
In this episode I talk about running an open weights model locally. Two main hardware lines: Apple Silicon and NVIDIA, both good.
The Great Inversion: Why Knowledge Work Is Collapsing into “Vibe,” Somatics, and Wisdom
Note: This post was written by Google Gemini. For over half a century, the social contract of the Western middle class was remarkably straightforward: exchange twenty years of rigorous cognitive…
Agile AI Development Lifecycle
A framework and methodology for transforming pre-AI software development lifecycle into AI-native development at scale.
Delete Everything, Keep Graph -Karpathy
Bookmarked: Delete Everything, Keep Graph -Karpathy
FreeToken: running models which are too big...
I ran a 35‑billion‑parameter mixture‑of‑experts model on my modest desktop using a hybrid of GPU VRAM and system RAM, thanks to a new model engine that activates just a few experts per token. The…
When Agents Call Agents
Calls to agents increasingly come from other agents, as general-purpose assistants route work to specialists. On the wire, these are ordinary tool calls, but if you zoom out and look at them…
Reimagining Homework in the Age of AI by Prioritizing Engagement With the Material
An approach to homework that prioritizes time spent engaging with material, documents the AI-assisted process, and tests understanding through live discussion.
Fast and Hard Code
One of the memes on Twitter is that “programming is solved now.” I’m not sure to what degree it is, but one thing is pretty clear: the act of familiarizing yourself with a language no longer matters…
The Agents Have Eyes Now
Until now, every agent I ran through Replaybook was blind. It could read the incident report. It could SSH into a broken NixOS machine, inspect services, edit configuration, and restart things. Once…
235 Billion Tokens Later: Auditing What The AI Agents Really Did
In the last post I presented my multi agent decompilation setup for MW2, almost framed as a scream for help. I received great suggestions that I will gradually implement over the next few days.…
An Agent Can't Offer an MCP Tool It Can't See
MCP discovery tells an agent how to authenticate, not what a tool does. When a tool sits behind per-user login, static metadata is the fix, not a smarter model.
How do you know if someone hates AI?
Don’t worry. They will let you know. And won’t they just. In a 2,000-word blog post. On a website they coded themselves. With a disclaimer in the footer that no part of it was generated,…
The Five-Minute Window
In March 2026, Anthropic silently reduced Claude's prompt cache TTL from one hour to five minutes — no blog post, no deprecation notice, no API version bump. For production workloads designed around…
Responding to Bala, again
Bala responded to my previous critique of his post over on Startup Iceland. In the same response, he responded to an article I posted on Vísir yesterday, where I responded to Carl…
AI Should Multiple your Team - It Shouldn't Replace Your Judgement
The organizations that work with AI will not be those that replace people the fastest. They will be the ones that use AI to multiply their teams' judgment, experience, and capabilities, without…
You don't have to make money with AI. You could just be happier.
I had two weeks of a Claude Max subscription left and tried to make it pay for itself. That made me anxious. Then I spent the rest rebuilding the tools I use every day, the way I'd always wanted…
AI Agents Take Over Lesson Prep
It has been a while since I have written about how I use Emacs/LaTeX to generate lesson notes for my students. Quite a lot has happened during that time. One of the main game changers has been the…
Tu quoque, Brute?
22 Aug 2026 Here’s another post where I wrestle with my ambivalence towards the technology I’m currently working with... 'Tu quoque to you Sir' OK, so one response to people pointing out generative…
kagent with agent substrate
After trying out kagent with a simple MCP server, I noticed that kagent supports substrate and I wanted to try it out to see how it behaves. Agent Substrate I started doing research about agent…
Atproto Is the Context Window You Own
I don't think it's a coincidence that atproto is taking off at the same time AI is. Well, okay, it kind of is a coincidence. I don't think anybody planned these things to work out at the same time.…
Agent Identity: What the Standards Fix, and What They Leave Open
SPIFFE, WIMSE and OAuth solve agent authentication. The three major clouds already ship most of it. None of them decides what an agent may do once it holds a valid credential, and the standards say…
Empire of AI (Hao, 2025, audio)
The AI Era (or, as I like to say, AI Crisis) can really be attributed to a single company: OpenAI. As such, it’s an important enough subject that I’m glad someone wrote a book about it…
Hold Our Own Ideas Loosely
I had a long conversation with Claude.ai. It was mostly about interview techniques for hiring software engineers, but as is my habit, it got philosophical. At one point I complained about timed…
Introducing SqlFlow: Building Resilient Workflows with SqlFlow (Python SDK)
This article shows how to build an Autonomous AI Agent with SqlFlow
Semionauts in latent space: Covert Design case study
Exploring what it what it means to find authorship inside an image generator's endless output.
Most of Your Architecture Was Just Expensive Code
For decades, architectural knowledge lived inside the implementation, because the implementation was expensive to reproduce. That expense is disappearing. And as it goes, we're going to find out how…
Younger People’s Pessimism Concerning AI is Growing
Colleen McClain and Eugenie Park at the Pew Research Center: Today, 52% of Americans say they are more concerned than excited about the increased use of AI in daily life – up from 37% in 2021.…
Trouble in ParadAIse
On the curious economics of a revolution with two paying customers Every gold rush has its shovel salesmen. The AI boom has improved on the formula. Here the shovel salesmen fund the prospectors, the…
An antidote to AI hype: AI agents are not “going rogue”
Dangerous, automated security tools existed before LLMs, and AI agents don't change cybersecurity fundamentals.
Turning a single work session into a reusable Skill with Microsoft's Skill Recorder
In my effort to adopt an 'AI first' philosophy, I spend a lot of time automating my day-to-day activities and transforming them to skills. It's not the most exciting activity: I know exactly how to…
Log Level Thoughts: LLM linting, 'wireless', other Mcluhan-isms
stick around to the end for an imagined job posting from a world where tokens are a first class consideration
Toward better mental models of learning with AI: AI as possibility machine
Cross-posted from my Intentional Teaching newsletter. One of the things I appreciate about designing and facilitating workshops for faculty is that doing so gives me the opportunity to take a few…
Data for Good! Vancouver AI Community Meetup on August 26, 2026
An August 12, 2026 notice (received via email) for Vancouver’s next AI community meetup promotes an upcoming August 2026 meeting on this event page, Taleeb Noormohamed, eh? This description of M.…
What ChatGPT Thinks of Me
I recently got a marketing email that suggested that I put this prompt into ChatGPT: Tell me about author Robert J. McCarter. What do they write? Who are their books for? What themes come up in their…
Evaluating Ox Alpha on Hack The Box Challenges
A new model called Ox Alpha landed on OpenRouter last night, and you can use it for free through Monday, August 24. Today, I tested it on my HTB-Challenger BenchmarkThe HTB-Challenger Benchmark…
Do not talk to your LLM, write to it!
Skip to content Home Sponsor me RSS bsky-comments { --background-color: var(--bgColor); --text-color: var(--textColor); --link-color: var(--textLinkColor); --link-hover-color:…
Recommended Reading: “To teach in the time of ChatGPT is to know pain” by Scott K. Johnson
Looks like I found another cool article I wanted share. This time it's about the impact of generative AI on education.
On how Jevons' Paradox related to AI coding is weird
On how Jevons' Paradox related to AI coding is weird
The Holdout Is a One-Way Door · Evidence cannot independently confirm the claim it helped shape.
The first cross-validation scheme in the audio project looked clean at the row level and failed inside each row. Each row contained a pair of performances, and the pair rows did not overlap. But an…
“AI will only do the drudging work,” they said
The counter to the perennial argument that "humans have seen their work automated before". https://filiph.net/text/2026-08-21-11am-no-we-haven't-seen-this-before.html
Delve
Generative AI has turned into the kind of arms race we normally associate with doping in sports. By which I mean: the companies involved make a legal product and a bunch of people try to figure out…
Self-playing games
Games that play themselves are like using AI for writing, drawing, and thinking, or, at the risk of making many people angry, watching others playing with balls on TV instead of playing yourself and…
UX Roundup: AI as Interviewer | Design Theater | The Planning Fallacy | Humans as Middleware | Long-Horizon Tasks | Good Videos
AI conducting user interviews mostly failed to probe deeply | AI UI tools break 1/4 of their design promises | The planning fallacy makes us believe more in schedules than they deserve | Cards as a…
Interesting stuff
I asked Claude lots of questions. These discussions are worth sharing: Compare CA, DE, NL and US on cannabis policy, prices, access (locals and tourists), crime and medical outcomes using data that’s…
AI Sparks Productivity Explosion in Maths
It may not seem like it, but the millennia-old academic discipline of mathematics has been undergoing a productivity revolution during the last thirty years. That revolution is being enabled by the…
Shift Left
AI coding agents have changed software engineering: when execution becomes cheap, the real leverage shifts to understanding and planning.
Migrating ts-loader to TypeScript 7.1 with AI
How I used AI to migrate ts-loader to TypeScript 7.1.