Building a small MCP server and wiring it into Vibe by hand, and my verdict after a semester of choosing Mistral: viable, sometimes the right tool, but a compromise.
An account of our trip to Engestofte Gods on Lolland, the feedback on our prototype, and why understanding the customer’s domain is the part no agent does for you.
What worked when my partner and I built the Engestofte project with agents from several providers: the plan we dropped, the design that made parallel work possible, and what model size really costs.
Reflections on spec-driven development and Naur’s ‘Programming as Theory Building’, the Pocock workflow I now use, and the start of our project for Engestofte Gods.
What I took from two days on LLM APIs: how Mistral and Anthropic want to be prompted, why every message costs you the whole conversation, and where skills fit in.