BloG

The unit of design is not the individual, and it is not the agentic system. It is the organization.

A framework for operating AI systems in expert-driven work

Enterprise AI deployments are accelerating. The people who need to operate, adapt, and own them are not part of the design conversation. This piece is on the maintenance philosophy gap at the heart of enterprise AI, and what choosing differently looks like.

The OpenTelemetry GenAI conventions have built a remarkable foundation for AI observability. This post describes a proposal to extend that foundation one layer further into the remediation loop, turning detected failures into measurable reliability growth.

Reins AI introduces a new blueprint for “Reliability & Repair” in AI, showing how complex, agentic systems can be continuously monitored, adapted, and improved in the real world.

You don’t need interpretability. You need direction. Reins AI evaluates AI the way we evaluate people: by behavior, not by wiring.

This paper proposes a method to measure cooperation using entropy over communication logs, validated on the Enron corpus, and designed to guide Cooperative AI evaluation.

A strategic overview of how AI deployment responsibility is shifting across enterprise segments, and what it means for AI safety efforts.

A visual guide to help teams design better multi-agent LLM systems—based on task planning, control structure, and agent ability.