One Agent Beats the Testing Agent Swarm

🔥 What happened In a heise practice report, Paul Behrend lays out a different play: instead of spawning many single agents, one central agent runs the test process under clear guardrails, loading task-specific skills for analysis, design, and test-case generation. 💡 Why it matters More agents doesn't mean better quality — just more cost and overhead. The catch: skill quality is tied to the underlying language model, and swapping models shifts answer quality noticeably. Classic automation still wins in some scenarios. ⚡ Our take Agent scaling is often just expensive randomness. Real test quality comes from one orchestrated workflow — not a zoo of specialists.