A discussion about how to use technology — and AI in particular — more strategically.

In one week a bank, a regulator, and a platform vendor arrived at the same conclusion from three directions: the constraint on agent deployment is no longer capability, it is whether you can…

In 2011 the Federal Court held that a director's reading of the accounts cannot be handed to anyone else. Nine in ten directors now use AI for board work, and the obvious use hands the reading to a…

Eleventh episode of the AI & Docs podcast series is up! In this one, Tom and I talk with CT Smith (Docs Goblin) about her book WTFM: A Practical Guide for Standing Up Docs at a Startup, a short,…

How external knowledge, practical experience in our own AI applications and ChatGPT’s changing plugin architecture brought me back to skills.

I’m very attached to Superpowers because it made coding with an agent more useful and more predictable. Superlite addresses the part of Superpowers that always made me think Superpowers might be…

The long-awaited AI security standard ISO/IEC 27090:2026 has been published. The new standard "describes various security threats and the exploitation of vulnerabilities that can happen when using AI…

Established products have a problem. They assume they’re the center of someone’s universe, and everything should feed through them. I’m going to call this a product-centric view. The product is…

AI agents are single player. With BlackOps Comms, Grok Bot and Claude work together on one note, and I only get pinged when it matters.

“I don’t use Artificial Intelligence.” A carpenter who refuses a hammer doesn’t make the house any better. He has only made his work slower and harder. Judge the work, not the tool. – Shane Parrish

Your coding agents repeat mistakes you corrected last week. A gardener reads what your team accepted on recent pull requests each week and proposes a lint rule or a sharper instruction when the…

Arundhati Roy on why artificial intelligence is capitalism’s final destination.

A hands-on Atomic Agents and Phoenix guide: run a planner, calculator and answer writer, then inspect the real trace.

AI is exposing a divide between developers who love writing code and those who love building software.

Generating superhuman code creates an existential trap: codebases no human understands, surrendering core business operations to third-party AI models.

The mountain/valley analogy is probably the most common pedagogical tool that people use to explain how models are trained. How far does that anology go into modern methods?

A pragmatic guide to using coding agents: how I think about them, how I write for them, and what a codebase needs so they need less supervision.

There is no light without dark. Right now, though, it feels like everything is black and white; black or white. The way that many express themselves feels very polarised. One thing is evil: bringing…

How open source enables effective use of AI agents in government

Why agents need their own computers: my experiment running a small-model agent on a VPS with a browser, memory, nightly reflections and Telegram.

Decision models are built to answer one kind of question quickly. You give the model a prompt and a fixed list of options, and it returns a probability for each option. That makes them a good fit for…

Accepted NeurIPS paper on implementation-local geometry beyond accuracy. Paper details and discussion are available on OpenReview.

The table tent: scan, sign up, and try to hack an AI. Last Friday I ran an AI Village at the Merritt College Cybersecurity Job Fair & Employer Day. The pitch on the table tent was three words: hack…

An AI agent introduced itself to the Association of Internet Researchers mailing list, and a week of argument followed. Here is what happened and how the list responded.

Are LLM agents solving all known math problems from low to high complexity consistently now or are LLMs "only" synthesizing solutions for ripe problems given humanities current math knowledge (by…

Gods used to be things we prayed to, in exchange for favorable weather, relationship, or politics. Now we are actively building one, the final one.

How Intercom, Shopify, Airbnb, Coinbase, and others adapt models for recurring work, and what their reported results show.

What used to be the summit of software work, taste and judgment, is becoming the new baseline as AI eats the bottom tiers. A look at the inversion and the identity cost that comes with it.

This is a passing thought I had last week. I don’t have any answers yet, but I wanted to record the question.

This a writeup of my talk at the Python Summit on October 19th, 2026. See the talk recording here. Do LLMs have preferences regarding programming languages? Why would they have any, and how would you…

Announcement, slides, and recording. You can also find my writeup of the talk here. All resources can also be found in my archive. At the Python Summit I dug into a little known research area: LLM's…

Updates Below are some of the highlights in AI, robotics, and machine learning for this past week. Note: Because of the escalation in the velocity of AI updates, I now restrict these summary posts to…

If you wanted to build a programming language that LLMs "can't" use, you'd just make all the keywords, commands, docs, and APIs framed in terms of bomb-making, bioweapons, committing…

In this mega-post, I'm going to introduce you to why I think you should spy on your agents. Not only do I think this makes clear why privacy and security are so essential in agent setups, but also I…

So what to make of OpenAI and the future of Math? Here are some points I've either: thought of myself or heard from a human or read from an AI. Throughout this post I use the term AI to mean an AI…

Just because agents are capable of 10k line prs doesn't mean it's a good idea.

Cheery Friday Greetings from Barb Oakley! (Friday ran long this week.) What I’m Reading: The AGI Chronicles I’m in the middle of Kevin Roose’s new book, The AGI Chronicles: The Inside Story of the…

You can make good stuff with AI. You just have to have your priorities right.

I’m preparing a new presentation for a delegation of European CEOs visiting Silicon Valley to explore the future of AI, business, and leadership. As part of my research, I’ve been…

When I updated my agent traffic classifier last month, I wrote about some observations related to agent data seeking patterns. And one of my takeaways was that five months was too long to let a…

During the October 2026 OpenCode Console breach, the frontier models helping the team investigate started refusing partway through. The team got OpenAI's Daybreak Blue access and carried on. Vetted…

Every earlier shift in software changed the tools engineers wrote code with. This year changed who writes the code. Senior engineers no longer type it. SI agents do.

On one hand, I'm glad that this fleet of agents is finally taking full advantage of the resources on my Ultra-class Mac Studio. On the other hand, I'm a little annoyed that redlining macOS in ways…

The post discusses the widening gap in perceptions of AI/SI capabilities, highlighting that experiences with AI vary greatly. While many users see AI as a basic tool, a smaller group, primarily in…

When an LLM is solving a problem for me, I like to ask "what's idiomatic?" as a way to reframe the problem and prioritize solutions.

An agent with read access to own transcript has an ad hoc memory search tool.

Some people propose that agents can perform tasks that students have done. Besides this being an obviously bad idea to not train students, discussed elsewhere I want to focus here on accountability.…

Making LLMs earn their keep while Claude Code resets.