the president is prompt engineer in chief pulling the Overton window this way or that speech for effect open source prompt injection or is it free programming we the people

Like many software developers, I’ve been considering the future of my profession in the face of AI.AI can write code. Our job is to write code. It seems natural that we’d be replaced by…

  For a while now I have been waiting for the "AI bubble" to burst. I don't want it to happen, but what the hell is one supposed to think when companies that make up a large portion of the U.S.…

In China, the pecking order is clearer. A founder who flies too high is grounded; a platform that forces monopolizing behavior and mistreats its workers gets punished; when the government rolls out…

Many prestigious scientific venues have suffered from papers containing hallucinated citations. GPTZero reports finding 100 hallucinations in NeurIPS 2025 accepted papers and 50 hallucinations in…

Qwen 3.8 is out since Friday, 17 of August. I am running it on my own server since Saturday. I am running Q4 quants from unsloth. So far it feels like a step up from Qwen 3.6 27B, which sometimes had…

In the year since initially building the foundations of the Data Introspection Project and building out a practice of personal archival and data tooling, I have been considering what artifacts and…

Although I do not suppose that either of us knows anything really beautiful and good, I am better off than he is – for he knows nothing, and thinks he knows. I neither know nor think I know. RLed…

    40 Bible Passages About AI: Scripture for an Age of Intelligent Machines releases tomorrow, and I wrote two of the chapters. Chapter 25 is on 1 Thessalonians 5:19-22, where Paul tells…

One of the consequences of LLM proliferation is that significantly more academic papers are uploaded to arXiv and submitted to conferences. As the quality of papers drops and their number increases…

Coding and research agents change the cost structure of quantitative hedge funds. But this does not provide for a particularly good opening for startup managers.

My friend John Gruber is incensed by the idea that AI watermarking schemes adjust the output of a given LLM so that the output can later be assessed as likely coming from it in particular: They say…

The Ask LukeW feature on my Web site has been answering people's product design questions using my writings, talks, images, and videos for over three years. During that time, I've seen people ask…

I’m seeing more instances of docs and README files addressing agents directly, as in “Hey, if you’re an agent, follow these instructions”. In some cases, those instructions are visible to human…

Introduction A plea to the undecided voter Table Notes Debian LLM GR - Summary of the options Introduction LLMs have finally made it to the ultimate stage of Debian’s governance processes, a…

“Here’s today’s batch,” he said gleefully, handing over a double-bagged parcel of half-brewed slop weighing exactly two pull requests and 11,700 lines of code. The batch will reach a…

Generate tailored review personas for any project. Reads the codebase, presents a role menu, and produces persona files with domain-specific expertise, review criteria, and voice.

In a previous post, I argued that AI is becoming a company operating system layer. The natural next question is how to tell whether that is actually happening inside a company. The answer is not how…

Gen AI, as you know, can produce polished steps faster than most of us can find the answer key. For students, this can make learning something new even trickier. Try the ACE framework.

The conversation about AI productivity is stuck on a number. 55.8%. 26%. 19%. Pick the percentage that supports your argument and you already have a slide for the next meeting. The problem is that…

A reader pointed out that in my second assignment, I sort of waved off the question of “What does validation look like?” My protagonist was, to be fair, quite snarky in telling the professor what…

Part two of UXR for AI-native teams, a six-post series.Last week I made the claim this series defends: at the speed of AI-native teams, the only research that matters is research built into how the…

AI slop isn't usually a bad model, it's bad context. Here's the discipline that prevents it, the failure modes behind it, and three checks you can actually run to enforce it.

This article argues that AI is collapsing the distance between vision and execution, ushering in a Doer Economy where the primary winners are those who can both imagine and build. The traditional…

The paper is older than Java. And it answers a question everyone is asking in 2026: Who guards the architecture when AI agents write the code? The post A Paper from 1995 Describes How You Should Work…

For as long as I can recall, there's been a 'thing' in parts of the UK – honesty boxes. As part of our (once) high trust society, people put their 'wares' – whether that be cakes,…

Tests are the place you can trust to see if your AI coding agents thought what you meant.

Epistemic status: Obviously speculative but mechanism design is fun. Last year I wrote a post on reward hacking as we were then beginning to see concerning signs of scaling RLVR causing models to…

Introduction: The Outages Aren't Bugs; They're Symptoms GitHub is down again? Oh boy, it has a scalability issue, you might say. I say your coding agent may be to blame, and thus may be you should…

You come back from a two week trip with forty hours of raw footage, and the editing alone eats up the next three weekends. That is the part of travel vlogging nobody warns you about when they see the…

I see you're reacting to all my messages with an emoji. It's to flag them and deal with them later using an LLM. You said AI first, right? You're reacting with a middle-finger emoji. The pin emoji…

“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 “if only you…

In the last 10 years or so, there has been a paradigm shift in how we train AI models. In particular, the best practice has shifted from Model-Centric AI, where we endless tweak model architectures…

Wiring telemetry and logging into the application and building the spec planning feature using the coding agent

Table of Contents Introduction: One Year of Self-Hosting AI The Motivation: Breaking Free from the Cloud The True Cost of Ownership: Cloud vs. Local SSD Upping the Game: Enter the NAS The 12TB…

A few days ago, a paper named “Stealing Reasoning Traces from Proprietary LLM APIs” was published. It describes a simple, yet super elegant way to recover encrypted LLM reasoning traces.…

When working with LLMs I find myself needing to update or correct the model with information specific to my work environment. The in-house processes, bespoke systems, and tribal knowledge of how…

I spent five months building a context engine on a premise I never checked. The premise seemed too obvious to test. An AI coding agent given a task starts by looking around: it greps for a symbol,…

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…

Last time when I started to write this post, I got carried away. My brain was on verge of explosion because Claude and I had worked together on so many projects and stuff over the last six months. My…

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,…

Anthropic announced Claude.ai will watermark every AI-generated text from now on. And online communities are having a complete meltdown over it. Freelance writers fear clients mistake human-written…

In Curated MySQL Data Sets for Realistic Testing I described datasets assembled the manual way — download, schema, load, validate, document — over hours or days per source. This post is the follow-up…

I was watching the (well made) video “Can We Trust AI to Code Without Human Oversight?” by Sam Newman on the Modern Software Engineering channel (recommended watch). He asks how much…

Most conversations about using AI for software engineering boil down to two extremes: breathless hype about autonomous agents replacing engineers by next Tuesday, or cynical dismissals based on an…

Thoughts on how the open source world is changing in the AI era, not always for the better.

The Best Way to Teach Students About AI Is to Build Something Together

Under a very specific but nevertheless very common reading of the practice of “tokenization”, tokenizers are things that turn a string into sequences of atomic identifiers, i.e., integers. These…