Building with an AI Coding Agent
A planned investigation of AI-assisted development under explicit repository constraints.
Question
How does an AI coding agent work within explicit product, architecture, and content constraints?
Proposed method
Use Blueprint itself as the independently created subject. Give the agent repository-owned specifications, a bounded task, and explicit quality gates. Review its changes against those requirements.
Record decisions, corrections, and verification evidence in the repository. Separate observed behavior from assumptions.
Evidence to collect
- Whether changes follow the approved scope.
- Which ambiguities require human decisions.
- Which tests reveal defects or missing requirements.
- Where human review changes the proposed implementation.
Current state
This experiment is planned. No comparative measurements, productivity claims, or conclusions are available. A later update may document actual observations, including failures.
EXP-001 / BLUEPRINT