- 2026
- 17 Jul 2026 Loop engineering: what the new way of working with AI agents is worth — and when a good prompt is enough
Loop engineering has AI agents work in cycles until a stop criterion takes hold. Which loop types exist, why loops fail in practice, and which seven questions to answer before your first own loop.
- 3 Jul 2026 Not every rule belongs in the prompt: steering AI instructions and model memory
Six zones decide where an AI instruction belongs — from the product settings to the current prompt. How to instruct ChatGPT and Claude cleanly and keep their memory of you current.
- 23 Jun 2026 The system prompt, taken from Claude Fable 5 — and which prompting techniques you can adopt from Anthropic
The complete 17,000-word system prompt was pulled from Claude Fable 5. What a system prompt even does, how far you can steer a model's answers — and eight techniques Anthropic builds into its models.
- 17 Jun 2026 GLM-5.2: the strongest open model is text-based and token-hungry
Z.ai released GLM-5.2 as open weights under an MIT license. In the independent Intelligence Index it leads the open field. Two caveats decide the practical value: it takes only text and consumes a strikingly high number of tokens.
- 1 Jun 2026 Where the tokens go: steering the context of AI models
The context window is finite, and the fuller it gets, the worse the model works. How to make the context visible, know its working zone and place a task into it without wasting quality and tokens.
- 10 May 2026 How Claude Code is built: memory, MCP, skills, subagents, hooks
Claude Code looks like a command line from the nineties. Beneath it lie six layers that make the tool a construction kit. A look under the hood, with examples from my work.