A research direction
Claude Code, Codex, Qwen Code, independent code audits, and multi-agent workflows are areas of interest. Comparative experiments are planned; no tool rankings or measured productivity gains are claimed. The one piece already working is a manual builder/auditor loop, written up as AITOOLS.
A proposed workflow
- Define the problem, constraints, and acceptance criteria.
- Inspect the existing code and operating assumptions.
- Implement a bounded change with reviewable differences.
- Run checks that exercise the intended behavior.
- Use an independent review to challenge assumptions.
- Review the integrated result and document limitations.
Multi-agent coordination
Parallel work introduces integration costs. Experiments should define ownership, avoid overlapping edits, and verify the combined result. More agents should not be assumed to improve every task.
Local and hosted models
Compare tools on equivalent tasks and preserve relevant configuration. Consider privacy, context handling, task success, and review effort alongside execution speed.
See the AITOOLS builder/auditor loop