Interpretability
What models reveal, hide, compress, and "think" beneath the answer.
Essays
AI is moving from tools we prompt to systems that plan, delegate, verify, and act. These essays examine what that shift asks of software, companies, labor, and human judgment.
Intelligence is becoming abundant. Trust, context, taste, accountability, and agency are becoming the scarce things.
Alignment
How agentic systems stay bounded, useful, honest, and accountable in messy real environments.
Societal Impact
How AI changes work, institutions, markets, education, and power.
Frontier Systems
Agents, enterprise infrastructure, cyber risk, autonomy, and self-improving workflows.
Latest essays
Full index
The Acceptance Gap
Cheap output creates value when someone can accept the result. Measure AI productivity across the whole workflow, including review and recovery.
Essays
The Cost of Leaving
Personal AI is portable when useful context survives a change of provider. An export button is only the beginning.
Essays
The Representation Gap
An agent can get the task right and still get you wrong. Personal intelligence needs a mandate its owner can inspect and change.
Essays
The Robot's Full Bill
A robot's purchase price is only part of its economics. The useful unit is an accepted physical outcome, with integration and recovery included.
Essays
The next internet will be built for agents discovering, verifying, negotiating, and acting on our behalf. The real test is whether we design that infrastructure with open protocols, accountable identity, and human agency at the center.