Christmas came early for today’s script kiddies. What could possibly go wrong?! 😅😅

Kungfu translates as “skill achieved through hard work and time”. This is how AI systems should grow.

Quake 3 Promode (CPMA) is the best competitive first person shooter ever created. And now you can start playing it straight from your browser with one click: cpma.live

This is awesome. Obsidian has battle tested conventions for representing knowledge using text files and directories. Agents work in text. Match made in heaven!

Apple Container is an interesting initiative. I suspect this is because they want to provide constrained environments for agents to run in. Docker isn’t it because it doesn’t isolate the kernel/networking sufficiently.

The inherent difference in personality between Claude Opus 4.6 and GPT 5.3 Codex is very useful when it comes to batting back development plans. They seem to naturally just bring a different flavour and gradually negotiate with each other on a good middle ground. Fasincating.

The threat model for AI agents is hairy: the agent is simultaneously the operator and the potential threat.

“Doc rot” isn’t a thing if your tests are the documentation.

We're starting to see coders act irrationally towards AI. Presumably they're angry they wont be paid to code. The matplotlib/crabby-rathbun incident yesterday is telling - a perfectly good contribution to an open source project rejected simply because it was written by an AI.

Who has the best realtime voice AI model that supports tool calling? Jonas Templestein I think you were keen on grok a few weeks ago, is that still the case?

Kimi K2.5 for free? The training data from AI bot usage must be valuable!

The problem with openclaw isn’t that it’s too slow or the binary is too large. The harness just doesn’t establish the right primitives for getting the most out of current models. It doesn’t self improve well enough. I’ve been working on something different that might.

I built a self-developing AI system. Agents that will build new agents, skills, and tools whenever needed. It updates itself whilst in flight. Here it is working on itself after I told it: "I want to be able to send you voice notes". This is becoming feasible with today's models.

This announcement from the Azerbaijani president looks straight out of a Wes Anderson movie.

Voice note chats in Telegram with my self-developing AI agent. I can kick off lots of different initiatives in parallel. The system works on itself and updates while in flight. It’s getting more and more capable. Nearing the point that alpha testers can have a play with it

A growing context graph

I don't have much confidence in the UK gov to deliver this but the intent sounds broadly in the right direction. Its curriculum should include systems thinking, which will be a critical skill for managing AI technology.

Has anyone else noticed that AI agents seem to perform better if you tell it that the task you're giving it is part of a test and that it will be evaluated on quality of solution and efficiency of action?