Why AI systems are failing in familiar ways

Multi-agent AI research reveals that software project failures stem from batch size and coordination complexity, not human factors. An experiment comparing multi-agent vs. single-agent AI setups showed that agent swarms perform worse due to compounding communication misalignment — the same structural problem that plagues large human teams. The fix is the same regardless of who writes the code: deployment automation, test automation, and monitoring to enable Continuous Delivery and small batch sizes. DORA research and Brooks' Law both support this conclusion: smaller batches reduce complexity and coordination overhead for both humans and AI agents.