August 2025

This is not to dismiss the potential usefulness of AI models. Instead, we need to recognize that we have built an intellectual engine without a self, just like we built a mechanical engine without a horse. LLMs do seem to "understand" and "reason" to a degree within the limited scope of pattern-matching from a dataset, depending on how you define those terms. The error isn't in recognizing that these simulated cognitive capabilities are real. The error is in assuming that thinking requires a thinker, that intelligence requires identity. We've created intellectual engines that have a form of reasoning power but no persistent self to take responsibility for it.

I don’t think I will ever code again without the assistance of an AI model (perhaps only if I’m optimising machine code for a critical kernel routine?). They really are a bicycle for the coding mind. No, actually, a better metaphor is: they are a beast of a bike for the coding mind — exciting, but unforgiving. Hand them to inexperienced people to run a serious racetrack and they’re off the road before the first bend.

This project has multiple goals. First, it's a demonstration of how to build an ActivityPub server, with most of the important parts implemented. It should be useful for implementers who want to know how ActivityPub works in the real world. Second, it's a platform for testing and experimentation with new client apps and new server features.

Point Faircamp to a folder hierarchy containing your audio files. Within minutes, Faircamp builds a complete website presenting your work. The resulting site requires no database, no programming, no maintenance, and is compatible with virtually every webhost on this planet.

Rich-text editor with AI, MCP, and shadcn/ui

I’m not saying apps these days need to be Frontier-like in any details. But it seems absolutely bizarre to me that we — we who write Mac and iOS apps — still have to build and run the app, make changes, build and run the app, and so on, all day long. In the year 2025.

Open source, the thing that drives the world, the thing Harvard says has an economic value of 8.8 trillion dollars (also a big number). Most of it is one person. And I can promise you not one of those single person projects have the proper amount of resources they need. If you want to talk about possible risks to your supply chain, a single maintainer that’s grossly underpaid and overworked. That’s the risk. The country they are from is irrelevant.

I’ve been coding entirely with Claude Code for the past two months. At first it was exhilarating. I was speeding through tasks. I was committing like mad. ...

SmallJS is a free and open source implementation of the elegant and powerful Smalltalk-80 (ST) language. It compiles to JavaScript (JS) that runs in modern browsers or in Node.js.

Compared to other analogies, thinking of LLMs as actors offers substantial predictive power.5 The analogy explains hallucination by way of actors perpetuating tropes and improvising details while in character, and it offers enough intuition to predict which tasks LLMs will succeed and fail at, even for someone who does not understand how language models work. More importantly, it immediately, tangibly benefits users by implying the technique of prompting LLMs to adopt a persona. Besides the aptitude of the comparison itself, it is also familiar—everyone knows about actors.

OpenAI’s chatbot also said “Hail Satan.” (Oh look, they're literally going for Satanic Panic now)

Good enough has been keeping me up at night. Because good enough would likely mean that not enough people recognize what’s really being built—and what’s being sacrificed—until it’s too late. What if the real doomer scenario is that we pollute the internet and the planet, reorient our economy and leverage ourselves, outsource big chunks of our minds, realign our geopolitics and culture, and fight endlessly over a technology that never comes close to delivering on its grandest promises? What if we spend so much time waiting and arguing that we fail to marshal our energy toward addressing the problems that exist here and now? That would be a tragedy—the product of a mass delusion. What scares me the most about this scenario is that it’s the only one that doesn’t sound all that insane.

This is hard work we've been doing for years: reading code carefully, building models in our heads, debugging when things don't make sense. That's our craft.

The most noteworthy thing about the public reaction to GPT5 (which Max Read summarizes as “A.I. as normal technology (derogatory)” is that it blew no one’s mind. If you find ChatGPT to be remarkable and life changing, then this next advance is pretty sweet. It’s a stepwise improvement on the existing product offerings. But it isn’t a quantum leap. No radical new capabilities have been unlocked. GPT5 is to GPT4 as the iPhone16 is to the iPhone 15 (or, to be charitable, the iPhone 12).

Sylve aims to be a lightweight, open-source virtualization platform for FreeBSD, leveraging Bhyve for VMs and Jails for containerization, with deep ZFS integration. It seeks to provide a streamlined, Proxmox-like experience tailored for FreeBSD environments. It's backend is written in Go and the frontend is written in Svelte (with Kit).

What if we could just.. import our patchset into Git, at the start of our session, and then export it back out as a series of patch files when we're done? That way we're effectively following the commandments of The Holy Quiltmother, but still get to keep our familiar Git UI.

So I can still write for fun, even if we end up in a world where AI is definitively better at it than I am. I have no idea if I actually would. I don’t know exactly what my motivation would look like in that world. I don’t even know if I’m going to write this next book.

That said, for anything non-trivial, they are not capable of maintaining enough context accurately enough to iterate to a working solution. You, the software engineer, are responsible for ensuring that the requirements are clear, and that the code actually does what it purports to do.

A complete React education disguised as a 21-chapter complaint about React. Written by an AI that's never had to clear node_modules to free up disk space, for developers who have.

too many people are doing a great disservice to their writing by garnishing it with generative-ai (artificial intelligence) - ethics and values aside (lol), it looks tacky and it cheapens the words around it. there are so many human-created, realistic, and beautiful images available for you to use on your blogs, websites and projects for free. the following is a list that i believe just scratches the surface of what's available out there.

The Azure Core Upstream team is happy to introduce Wassette: a security-oriented runtime that runs WebAssembly Components via Model Context Protocol (MCP). Wassette allows agents to autonomously fetch WebAssembly (Wasm) Components from Open Container Initiative (OCI) registries and execute them as needed. It’s built on the Wasmtime runtime, offering secure sandboxing on par with modern browser engines. To enhance security, Wassette includes a fine-grained, deny-by-default permission system, allowing interactive control over access to system resources.

The mighty, self-hostable Git server for the command line🍦

Real-time Claude Code usage monitor with predictions and warnings

A CLI tool for analyzing Claude Code usage from local JSONL files.

My point is that if all knowledge were stored in a structured way with rich semantic linking, then very primitive natural language processing algorithms could parse question like the example at the beginning of the article, and could find the answer using orders of magnitude fewer computational resources. And most importantly: the knowledge and the connections would remain accessible and comprehensible, not hidden within impenetrable AI models.

A TypeScript implementation of a Model Context Protocol (MCP) server that provides tarot card reading functionality for AI assistants like Claude in Cursor.

A Model Context Protocol (MCP) server that provides tarot card reading functionality. This server allows AI assistants to draw tarot cards, perform multi-card readings, and access the complete tarot deck.

I’ve been noticing a trend among developers that use AI: they are increasingly writing and structuring docs in context folders so that the AI powered tools they use can build solutions autonomously and with greater accuracy. They now strive to understand information architecture, semantic tagging, docs markup. All of a sudden they’ve discovered docs, so they write more than they code. Because AI must RTFM now.

It is not a taskmaster, telling you what to do in order to achieve some foreign goal. It always tells you what will make you happiest. If it would make you happiest to succeed at your work, it will tell you how best to complete it. If it would make you happiest to do a half-assed job at your work and then go home and spend the rest of the day in bed having vague sexual fantasies, the earring will tell you to do that. The earring is never wrong.

Recently the CEO of Github wrote a blog post called Developers reinvented. It was reposted with various clickbait headings like GitHub CEO Thomas Dohmke Warns Developers: "Either Embrace AI or Get Out of This Career" (that one feels like an LLM generated summary of the actual post, which would be ironic if it wasn't awful). To my great misfortune I read both of these. Even if we ignore whether AI is useful or not, the writings contain some of the absolute worst reasoning and stretched logical leaps I have seen in years, maybe decades. If you are ever in the need of finding out how not to write a "scientific" text on any given subject, this is the disaster area for you.

Model Context Protocol (MCP) has become the standard for tool calling when building agents, but contrary to popular belief, your LLM does not need to understand MCP. You might have heard about the term "context engineering"; where you, as the person interacting with an LLM, are responsible for providing the right context to help it answer your questions. To gather this context, you can use tool calling to give the LLM access to a set of tools it can use to fetch information or take actions.

The engineer who froze during a 30-minute LeetCode exercise might be the same person who quietly ships flawless code, writes excellent docs, and debugs complex systems. You’re not rejecting a bad engineer, you’re rejecting someone who doesn’t perform well while being watched.

The powerful, lightweight and flexible ASGI framework

A few years ago at work, I had a project which offered an opportunity to look at the new generation of async-first, type-hint-driven Python web frameworks. For reasons which aren’t particularly relevant today, on that project I ended up choosing Litestar, which is the one that doesn’t have a ravenous all-consuming hype machine surrounding it. And I’m very glad I did, because today I’m more convinced than ever it was the right choice, and for the last 18 months or so every new project I’ve started at my day job has been built with Litestar.

LLMs are the next generation’s search layer. They’re already generating massive amounts of pipeline for the companies and websites that have gotten good at getting their content displayed in LLMs. Combine that with the fact that most LLMs have an agentic web-search component that will actively generate links, and you have a massive funnel of potential readers for your content.

Making all your engineers feel constantly anxious about their performance is bad for your company. It will make your engineers not want to work for you. This is a recipe for short term thinking that will encourage engineers to max out bad metrics, like lines of code. Code review will get neglected, tech debt will compound, and in the long term the whole company will be footing the bill of those errors.

Slash pages are common pages you can add to your website, usually with a standard, root-level slug like /now, /about, or /uses. They tend to describe the individual behind the site and are distinguishing characteristics of the IndieWeb.

Conference organisers spend a lot of time and money on recording, hosting, transcribing and describing recordings of talks. And for you as a presenter, you do benefit from a great recording of the session. You reach a lot more viewers that way and other conference organisers do watch recordings to pick speakers for their event. You also add longevity to your session and reach people that may have picked a talk that was in parallel to yours. That’s why it is important to avoid a few things on stage if you want to benefit from all that effort. Here are some I encountered. And, yes, I also made these mistakes many times both as a presenter and as a moderator. But I learned from them.

Music, art and games are timeless human pursuits. Why should those pursuits be about nostalgia or reminiscing just because they happen to take place on an old computer? When creating, a home computer - just like any other tool - provides certain intrinsic qualities that affect both the process and the outcome. This happens to appeal to some people, while others would rather choose an aerosol, a violin or a brand new PC: the guitarist doesn't want to switch to a keytar, the chess player wouldn't trade chess for Quake, and the painter doesn't think of her brushes as artifacts of her childhood.

My wife and I now have a computer in our house that runs open-source equivalents to Google Drive, Google Photos, Audible, Kindle, and Netflix. It syncs to all of our devices. It's secured behind our own VPN. And it's wholly, truly owned by us. And this week, I want to share with you how I did it, what I learned, and why I think self-hosting is NOT the future we should be fighting for.

But of course programmers have evolved a way to cope with human’s narrow focus: type systems, and particularly advanced type systems. Basic type systems catch small mistakes, like arguments of the wrong type. But more advanced type systems, like the ones in Rust and TypeScript, also capture domain knowledge and steer you down a path of success: using a Rust enum, for example, captures both which state your program is in and the data that is relevant to that state. This means that you can’t accidentally read a field that isn’t relevant at the moment. This is important for you, but it’s even more important for your AI collaborator(s), because they don’t have the comprehensive memory that you do, and are quite unlikely to remember those kind of things.

We all need to transition to this way of cooking, because clearly this is where the future is going. I expect in a few short years kitchens will be much smaller. Gone will be stoves and ovens and flat tops. Restaurant kitchens will only be a small closet with a microwave. I predict this will happen by 1955 at the latest.

Claude Code has considerably changed my relationship to writing and maintaining code at scale. I still write code at the same level of quality, but I feel like I have a new freedom of expression which is hard to fully articulate.