Topic

Agentic AI

Building LLM systems that survive production — orchestration, grounding, guardrails, evaluation and the cost of running them.

The thread running through most of the recent work: agents that can be governed, audited and paid for, rather than demos. It shows up as an orchestration runtime with the guardrails in the kernel, as knowledge bases an agent maintains with provenance, and as measuring what an AI feature actually costs.

Principles