weekend ai reads for 2026-06-26

It came to my attention this morning that you appear to be using some kind of agentic AI system to try and resolve Fedora bugs. It’s great that you’re trying to fix things, but the results seem to be kind of erratic. I’m still working through your Bugzilla history, but so far I’ve seen several issues.

We are living in a society where cops kill children over a pack of stolen diapers.

And that breaks my heart, as it should.

If you have enough nieces and nephews, you understand that people are just themselves; they come out that way.

Look, this is complicated. We understand that people have strong feelings about generative AI and its impact on creative work (amongst other things), and the ethics of using it at all. Equally, though, it’s true that there are a significant number of ways in which generative AI can be used to help people make and build things online which they wouldn’t otherwise have been able to do—which feels like a good thing!

I’ve seen people joke about adding typos to emails to prove that they wrote them. MS Paint-style image macros read as more human than detailed, funny images (the image could be AI slop). Websites that look intentionally bad are more interesting than websites that look beautifully bland.

Conveying our humanity in the face of LLMs is a hard, new, interesting problem. I’m interested to see what we produce as a result.

Maybe it’s worth backing up for a moment: what is an AI agent? At its most basic, it’s a ‘while loop’ – a bit of code that asks an LLM what to do next, then does it. (Usually ‘what to do’ means using a program – called a tool – that can do things like search the web or send emails.) The tool does something, the result of doing that thing is fed back into the LLM (along with everything that came before), and it all starts again.

Verifier’s rule: The ease of training AI to solve a task is proportional to how verifiable the task is. All tasks that are possible to solve and easy to verify will be solved by AI.

Over the past year, every piece of my work that I could hand off, I handed off to AI, piece by piece. The design doc — it wrote. The code — it wrote. The first drafts of documents and review comments — it wrote. What I can now run in parallel in one evening would have taken a full quarter two years ago. By rights I should be idle, but the truth is the opposite — I am busier than I have ever been. The content of “busy” has simply changed: I almost never produce anything with my own hands now. I spend the whole day reviewing what it produces.

In practice, a handful of workflows drive the majority of your bill: an automation nobody is watching, someone who accidentally left fast mode on, a feature provisioned on the frontier out of caution. You do not need to optimize everything. You just need to find ten lines to solve 80% of your problem.

paper cups did not win because they were cheap. they won because the maintenance cost of the alternative -- washing, sanitizing, tracking down a cup that walked off -- went from “annoying” to literally spreading disease.

software is in the same moment. legacy code was never loved. it was tolerated because the alternative was expensive. now the alternative is a prompt.

Pupils from first ​through seventh grade, aged 6 to 13, should as a general ​rule not be using AI, while those in lower secondary school, aged 14 to ‌16, can ⁠cautiously adopt tools under teachers’ supervision, the government said.

In upper secondary education, from ages 17 to 19, students should learn to use AI appropriately so that they are prepared for further education and work, it added.

Refine, an AI-powered system that harnesses frontier AI models to perform substantive technical diligence on research work, won 90% of head-to-head matches against other LLM review systems on 150 economics preprints.

The scariest part to me is that we become dependent on these new machines in new ways. Software has always depended on tools. I remember the time when I had to pay for compilers. These new tools are a flashback to times where creating software came with real costs. But now it’s no longer a one-time payment, it’s a constant dependency. Not just a dependency on a filled wallet, but also a cognitive dependency.

What AI systems value increasingly shapes decisions in the economy, national security, and science. We measured those values across people, companies, and countries, and found clear differences in who AIs favor, trust, and want to see succeed.

Find out if you live on in GPT-5.5, GPT-5.4 Mini, Opus 4.8, Haiku 4.5, Grok 4.20, Gemini 3.1 Lite, Kimi K2 0905, DeepSeek V4, Llama 3.3 70B, Llama 3.2 1B, GLM 4.7 Flash, Mistral 3.2 24B, and Qwen3 8B.

Pokémon SVG Bench evaluates a model's knowledge of Pokémon and its ability to generate SVGs.

Guadagnino hasn’t publicly commented on the distribution challenges beyond acknowledging the Amazon exit. The director’s representatives are presumably working to finalize a deal with remaining suitors. But regardless of where ‘Artificial’ lands, the damage to Hollywood’s reputation for editorial courage is done.