CS-14 — Automation & AI
Development and auditing workflows with several AI agents
Experimentation with software development workflows in which one agent implements and another audits, iterating until the criteria are met, with a human in the loop.
- My role
- Personal and professional experimentation.
- Period
- Not documented
- Disclosure
- Public project
- Technologies
- ClaudeGitHub CopilotHerdr
- Last reviewed
- 13 November 2026
Context
Coding agents speed up implementation, but they need review. The idea is to separate implementation and review responsibilities.
Challenges
- 01Detecting defects, vulnerabilities, architecture problems or missing tests in AI-generated code.
Solution & architecture
An implementation agent investigates the problem and proposes and develops a solution; an independent auditor agent analyses the implementation and identifies defects, vulnerabilities, architecture problems or missing tests; the implementation agent fixes them; the cycle repeats until the criteria are met.
Key decisions
D-01
Separate implementation and auditing into different agents
A reviewer that has not written the code does not inherit its assumptions.
Trade-offs
Not documented yet
Results
Results will be published once verified
Lessons learned
Not documented yet
This is a line of experimentation, not a production-ready application. Ander has explored agent tools and workflows with Claude, GitHub Copilot and Herdr.
Ask about this case study
Questions here are answered with the case study “Development and auditing workflows with several AI agents” as context.