- Mechanistic Interpretability Is Governance, Not Just Debugging
A field guide for treating model internals as civic infrastructure: not mystical transparency, but inspectable constraints.
- Projection Mapping and the Return of Public Computation
Projection mapping turns architecture into an interface, and reminds technologists that computation can be public, situated, and theatrical.
- From Crisis Maps to Cell Atlases
The same pattern language behind geospatial intelligence for crisis response can help us reason about gene sequencing and cellular neighborhoods.
- AI Workflows Need Institutions, Not Just Agents
Agentic software becomes useful when it carries memory, accountability, review, and handoff instead of only speed.
- Forward-Deployed Engineering for Agentic Workflows
Why the Palantir-style forward deployed software engineer model matters for AI agents: context, field contact, systems thinking, and responsible branching workflows.
- Peer-to-Peer Software After the Cloud
Pear, Hypercore, Hyperswarm, and the return of peer-to-peer software as a serious systems pattern for resilient apps, local-first data, and user-owned infrastructure.
