Project
Prism
A task-driven, spec-governed workspace for AI coding agents.
Prism is the product idea I keep returning to when I think about where coding agents still fall short. The deeper problem is not raw code generation. It is task legibility: constraints get lost, execution state becomes fuzzy, and completion drifts toward vague intuition instead of explicit verification.
The current framing is that Prism should begin as a task contract layer, not as a full branching or adjudication workspace. That keeps the first version attached to the most common failure mode in AI coding workflows.
Current stage
Early product framing. The immediate question is how to turn a loose coding request into a live task contract that can anchor planning, execution, and verification.
Project notes
- Prism V1 — public note on why the first wedge should be the task contract layer.
- Private memo — internal note with the fuller framing, risks, and layer model.