A practical look at harness engineering through our daily OpenCode setup and a small Python agent called Sokrates. more
Gen AI
Harness Engineering for Reliable AI Agents
AI Model Rankings and Agent Harnesses in September 2026
What OpenRouter traffic says about the AI market: cheap fast models take the token volume, benchmark leaders take the headlines, and ten harnesses burn most of it. more
Why Laravel, Yii and Rails Suit Agent Driven Development
Conventions, good docs and little code: why fullstack frameworks shrink the context an AI agent needs, and why Django is not my pick. more
How I Lost 900 PLN on Gemini Tokens in a Subscription
A bill of nearly 900 PLN for Gemini API despite billing alerts. Two causes: a changed Google account email and paid subscription overages. more
Fast Code Generation with Agents, Rails, and MongoDB
Get useful code out of AI agents faster by pinning versions, keeping a short README as the source of truth, and sending small prompts. more
RAG in Practice: Anti-Patterns, Best Practices, and Why Simplicity Wins
Seven RAG anti-patterns that quietly ruin retrieval quality, and the simpler practices that fix them: own the context, structure over size. more
Multimodal RAG for Images - 2026 Guide
Build image search that answers in text: pick an embedding model, a vector database and a multimodal LLM, then cut the running cost. more
Why Law Firms Need Their Own LLM/RAG
Weigh a private LLM/RAG for a law firm: hallucination risk, attorney-client privilege, running cost and what implementation demands. more
My First RAG
How a 2019 visual product search was built, from crawling shop catalogues to vectorising photos and serving nearest-neighbour results. more