Featured post
AI harnesses: the missing layer between a smart model and useful work
An AI model is not an AI worker. The harness is the operating layer that gives it tools, memory, permissions, and proof.
Learnings12 min read
Read postThe notebook
Thinking out loud. Working things out.
Articles on AI architecture, automation, and the engineering choices behind them.
More from the notebook
Insights3 min read
The "Vibecoding" Trap: Why Your AI-Built App Is Probably Wide Open
In an experiment with ten people I know, I found security flaws in seven of nine projects built with AI; working software was not necessarily secure.
Read postLearnings8 min read
Scaling n8n with queue mode and Docker: architecture and example review
Main, workers, Redis and PostgreSQL: a guide to queue mode architecture. The runner configuration needs review before the examples are used.
Read postInsights2 min read
Google wants to end the “15 open tabs” shopping struggle
Google presents UCP as a way to shop through assistants. This analysis of the January 2026 announcement distinguishes that promise from actual adoption.
Read postLearnings3 min read
Prompt Caching: A Technical Guide to LLM Efficiency
Prompt caching can reduce the cost of reusing context, but the benefit depends on repeated input and each provider’s conditions.
Read postLearnings6 min read
AI Guardrails: Why Your Chatbot Needs a Safety Net
Guardrails help limit dangerous responses and actions, but cannot guarantee a safe chatbot: layered controls, testing and oversight still matter.
Read postOpinion2 min read
AI Agents vs. Chatbots: One Talks, the Other Actually Does Stuff
An agent can use tools to act on a task; moving beyond conversation also raises questions about permissions and oversight.
Read postLearnings10 min read
n8n 2.0: changes and context for the 2025 migration
A guide to the n8n 2.0 changes announced in December 2025. Check the documentation for your version before applying the examples.
Read postInsights3 min read
Claude Opus 4.5 in Simple Terms
A plain-language look at the announced Claude Opus 4.5 features and their potential everyday uses, in the context of its launch.
Read postInsights7 min read
Google's Gemini 3: "World's Smartest AI" (Spoiler: It Can't Even Tell Time)
A critical look at the November 2025 Gemini 3 launch: advances, limits and the need for supervision. Includes a correction to the cost calculation.
Read postInsights3 min read
I Made a Mistake About TOON (And Here's the Data That Proved Me Wrong)
I correct my TOON recommendation: token savings need to be evaluated alongside accuracy, data structure and the model being used.
Read postOpinion3 min read
Are LLMs Becoming Idiots?
I worry about uncritical AI content finding its way back into training data: a reflection on source quality, not proof that every LLM is getting worse.
Read postLearnings3 min read
TOON vs JSON: token savings and limits of the original recommendation
The original TOON recommendation was qualified in a correction: saving tokens does not guarantee unchanged accuracy. Read both pieces before switching formats.
Read postLearnings3 min read
n8n Introduces Project-Scoped Variables: A Game-Changer for Team Collaboration
Project-scoped variables separate configuration in n8n; this article explains their scope and how they differ from global variables.
Read postOpinion4 min read
Microsoft's "Humanist" Superintelligence: When Marketing Meets Existential Dread
A critical look at Microsoft’s “humanist superintelligence”: specialist AI can make sense without turning its marketing into a philosophy.
Read postOpinion5 min read
Companies have discovered that AI costs money
A satirical look at businesses discovering AI return on investment: higher spending and reported gains do not, by themselves, demonstrate success.
Read postLearnings3 min read
DeepSeek-OCR: The AI That Reads Too Much Into Everything
DeepSeek-OCR proposes compressing text into visual information, but fewer tokens must be weighed against the risk of lost accuracy.
Read post