The Specification Surface Is the New Source of Truth
A successor flagship essay on why AI-assisted software work becomes governable when intent moves from private prompts into maintained, testable specification surfaces.
Public research mirror
Substack is still the canonical publication surface. This mirror keeps the public essays close to the repo folders, raw markdown snapshots, and source archive that produced them. The point is not redundancy. The point is legibility.
Featured essays
A successor flagship essay on why AI-assisted software work becomes governable when intent moves from private prompts into maintained, testable specification surfaces.
A flagship synthesis on how AI workflows borrow trust when polished artifacts, evaluation scores, and human approvals move faster than their evidence.
An investigation into why LLMs often preserve semantic ingredients while failing at the exact relational binding that makes the answer faithful.
A research-backed argument that many multi-agent systems are temporary routing workarounds rather than the final architecture.
Series
7 mirrored posts in publication order.
4 mirrored posts in publication order.
Recent additions
A successor flagship essay on why AI-assisted software work becomes governable when intent moves from private prompts into maintained, testable specification surfaces.
A flagship synthesis on how AI workflows borrow trust when polished artifacts, evaluation scores, and human approvals move faster than their evidence.
An investigation into why LLMs often preserve semantic ingredients while failing at the exact relational binding that makes the answer faithful.
The framing note for the Operating Agents course, introducing the chapter structure, blueprints, companions, and builder-first doctrine.
The optimization and evaluation bonus chapter of Operating Agents, focused on workflow optimization, measurement, and bounded self-improvement.
The safety chapter of Operating Agents, covering trust boundaries, attack surfaces, and architectural containment.