I Ran a 165-Agent Security Review on My Own Codebase. Here's What I Learned

Overview:

A first-person walk-through of running 165 Claude agents in parallel against a real codebase. We cover the orchestration setup, prompt design for adversarial review, triage of agent-flagged findings, the categories of issues that surfaced (and the categories that did not), false positive patterns, and the cost economics of running review at this scale. We close with where this approach belongs in a security program and, just as important, where it does not.

Why you should Attend:

Your security review process scales linearly with your team. Threat models miss what no one has time to look for. Meanwhile, attackers automate. If you have not stress-tested what 100+ agents can find in your own code, someone less friendly is doing it for you.

Areas Covered in the Session:

• Why scale changes what you find: parallel coverage versus sequential depth
• Designing prompts for adversarial review, one focused agent at a time
• Orchestrating 100+ agents without losing the thread
• Triage: signal-to-noise on agent-flagged findings
• Categories of issues human review reliably misses
• False positives, hallucinated CVEs, and how to filter them
• Reproducibility: re-running the full review on every pull request
• Cost math: what 165 agents actually costs per review
• Tooling: what to build versus what to buy
• Limits: where this approach should not be your last line of defense

Who Will Benefit:

• Application Security Engineers
• Security Architects
• CISOs
• DevSecOps Engineers
• Penetration Testers
• Security Researchers
• Staff Engineers
• Engineering Managers
• Heads of Platform Security
• Cloud Security Engineers
• Red Team Leads