Selected lines of inquiry3 inquiries

Projects

Public explorations of the systems questions I keep returning to. The emphasis is on architecture and lessons, not proprietary implementation.

01Active inquiry

Autonomous operations

Rollout Reviewer

A systems-design exploration of autonomous production review: incremental evidence, checkpointed decisions, explicit uncertainty, bounded authority, and outcome reconciliation.

Questions under investigation

What evidence must exist before a rollout can be called healthy?
How should a long-running review own time, state, and escalation?
Can decisions improve across applications without flattening local context?

02Architecture notes

Agent infrastructure

Universal Agent Harness

A model-driven harness where agents assemble workflows dynamically while tools, sandboxes, policies, identity, and observability retain stable responsibilities.

Questions under investigation

What should remain static when the agent graph is dynamic?
How should skills discover capabilities without hard-coded tool names?
Where do runtime state and accountability live?

03Evolving framework

Principles

Trustworthy Autonomy

A practical framework for systems whose consequential decisions are evidence-backed, uncertainty-aware, constrained by authority, recoverable, and auditable.

Questions under investigation

Which trust properties belong in prompts, policy, or runtime?
How should rigor scale with consequence?
What makes a decision independently verifiable?