# Insights

Articles from the Velais team on AI evaluation and the Model Context Protocol.

- [The hidden work behind good AI](/insights/hidden-work-behind-good-ai.md), 2025-10-26: AI products rarely fail because the models are wrong—they fail because nobody defined what right even means. Here's why disciplined evaluation turns demos into dependable systems.
- [You don't need a smarter model. You need a measurable system.](/insights/you-dont-need-a-smarter-model.md), 2025-04-22: Enterprise AI doesn't collapse because models are weak—it collapses because teams never measure how retrieval, generation, and cost behave once the system meets reality.
- [Microsoft Joins MCP: The SaaS Extinction Event Has Begun](/insights/microsoft-joins-mcp-saas-extinction-event.md), 2025-03-29: Microsoft's support for MCP signals the beginning of a new era where agents replace traditional SaaS apps. Here's why CRMs, CDPs, and integration tools should be worried.