Technical essays and architecture notes from building dependable AI and distributed systems. Rigorous focus on stateful multi-agent execution, durable checkpointing, model risk management, and token economics for 2026/2027 enterprise deployments.
A practical design for reviewable agent workflows: typed state, explicit approval interrupts, durable checkpoints, and idempotent resume paths.
A practical operating model for governing generative AI: inventory, risk tiers, evidence, change control, and monitoring tied to real system boundaries.
How durable checkpoints, explicit state, and idempotent effects make multi-step agent workflows recoverable and easier to operate.
A systems view of governed retrieval: data lineage, access-aware retrieval, citation checks, evaluation baselines, and production evidence.