Head of AI & Development
Michal Vávra leads AI and engineering at Pixelfield, where he takes AI systems from proof of concept to production. He writes about applied AI, LLM architecture, and the practical trade-offs of shipping intelligent software.
Model Context Protocol (MCP) explained for the people who buy and oversee AI: what it is, why it matters now, when to leave it to your vendor, and the security questions to ask before AI touches your data.
Frontier models made vibe-coding fast, but production is a different job. When DIY stops and an AI software development company should own security, evaluation, observability and cost control.
Why most multi-agent systems quietly fail in production. The reliability maths, the 4 failure modes, the 80/20 scaffolding split, and when to skip multi-agent entirely.
