The deadline moved. Your risk didn't.
The EU is postponing the AI Act's high-risk obligations by up to 16 months. Why guardrails, traces, and evals still don't belong back in phase 2: the delay pauses neither liability nor GDPR nor MDR.
Blog
Architecture notes
Short, practical writing on AI architecture, platform delivery, and the work that turns pilots into production systems.
AI liability is an architecture problem
Why chatbot output in customer-facing processes can't be treated as a side effect: an OLG Hamm ruling shows that guardrails, source grounding, human-in-the-loop, traces, and evals don't belong in phase 2.
The gap between agentic AI and enterprise reality
What roughly 40 self-built tools reveal about the real AI adoption gap: the bottleneck isn't the model, it's problem understanding, ownership, governance, measurability, and controlled execution.
An architect, but hands-on: what an AI Solutions Architect actually does
Not a role taxonomy, but the concrete work: start from the workflow, decide build vs. buy, draw data boundaries, build the MVP, integrate with the real organization, and ship to production.