Objective
Explore how AI coding tools can support implementation, troubleshooting, and code review while keeping changes understandable and testable.
Project status
This is a candidate experiment. No comparative results or completed workflow claims are published here.
Questions to investigate
- Which tasks are best suited to hosted and local coding models?
- Does an independent code audit identify defects the implementation assistant missed?
- How does task scope affect the usefulness of multiple agents?
- What makes a change easy for a person to inspect and maintain?
Proposed approach
Use small repositories with reproducible tasks. Give each tool a clear brief and equivalent starting conditions. Preserve prompts, patches, tool versions, and validation results after removing any private information.
For multi-agent experiments, separate file ownership or use isolated worktrees. Review integration changes explicitly instead of assuming individually successful tasks will combine correctly.
Validation plan
Evaluate correctness, introduced defects, review effort, and reproducibility. A passing test suite is useful evidence but does not by itself prove security or completeness.
Expected output
Publish a documented workflow and examples only after running the experiments. Avoid rankings based on a single task or undocumented impressions.