Notes on scaling work with AI agents
For engineering and marketing teams that want more output from the people they already have.
Claude's text watermark: who can detect it, and can you remove it?
Luka Breitig
Claude now watermarks everything it writes. I read Anthropic's write-up, the EU law behind it and the research to find out who can detect the mark and what removes it.
Read the article
Do LLMs write better Elixir than Python?
A Tencent benchmark ranked Elixir first for LLM code generation. As an Elixir developer, I checked the data and found the ranking overstated and the structural advantages real.
Read the article
From chatbot to agent: a step-by-step model of how AI agents get work done
Step by step, how an AI agent plans a task, uses tools and checks its own work, and how to tell which of your team's tasks it can take on.
Read the article
When your AI turns against you: defending public chatbots from prompt injection
Users turned Bild.de's public chatbot against its own brand. I walk through the prompt-injection defences that keep a public chatbot working for you.
Read the articleWant AI agents doing real work in your team?
Three ways to get more done with the people you already have.
- Training Teams learn to hand real work to AI agents, on their own tasks and tools. See how team training works
- Consulting I look at how your company works and show where AI agents add output, in any department. See how an assessment works
- Development Agent systems and custom software in Elixir and Python, built and shipped for your team. See how development works