The Cost of Usable Intelligence
A framework for measuring the cost of useful AI output under constraints like latency, reliability, safety, and quality.
Research
AI infrastructure, observability, and compute systems research.
Research notes, papers, and frameworks on AI infrastructure, silicon observability, autonomous systems, compute economics, and quantum-classical systems.
Featured series
Observability for Frontier AI Systems
This series translates lessons from silicon bring-up into frontier AI infrastructure: if complex systems cannot expose state, localize failures, and produce trustworthy traces, they cannot be operated safely at scale.
Silicon bring-up
Agent telemetry
Eval infrastructure
- Published Debuggability for autonomous agents: what AI safety can learn from silicon bring-up
- Published Observability is an underbuilt safety primitive for frontier AI systems
- Framework AI agents as infrastructure risk
- Research note A taxonomy of autonomous agent failure modes
- Lab-backed Building an Eval Harness for Cyberphysical AI Risk
Observability for Frontier AI Systems
A series translating lessons from silicon bring-up into agent telemetry, safety instrumentation, and eval infrastructure.
Observability is an underbuilt safety primitive for frontier AI systems
Agent observability stack, maturity model, silicon-to-agent mapping, and operational eval agenda.
A taxonomy of autonomous agent failure modes
A public taxonomy for failures that emerge when agents gain tools, memory, planning loops, and cyberphysical reach.
Building an Eval Harness for Cyberphysical AI Risk
A proposal for observable, bounded, and reproducible tests of agent actions that touch real systems.
Quantum computing needs an accelerated classical control plane
Why quantum systems will bottleneck on classical orchestration, observability, latency, and platform integration.
Debuggability for autonomous agents
What AI safety can learn from silicon bring-up: state capture, replay, failure localization, and root-cause discipline.