michelangelo.codes
Writing on the harder edges of cloud and AI, currently the seams between agents and what it means to keep workloads under European control.
Recent writing
-
Nobody Agrees What Delegation Means, and That Is a Security Problem
Three of the largest companies in the field have shipped three structurally incompatible answers to what it means for one agent to hand work to another, and all three sit under the same category heading. Choosing a managed runtime means choosing a trust model. The design for measuring what breaks at that boundary, on the record before any results.
-
The Agent Stack in 2026: Which Layer Is Actually Worth Standing On
The model is swappable, the harness is rentable, the protocols are open. The value doesn't sit on any one layer — it leaks upward, faster than anyone is building businesses on it. A field guide to where it's heading next.