Lucas Costa's article on building systems for code-generating robots sparks an interesting debate about metaphors. Costa frames the challenge as backpressure—signaling upstream processes to slow down—but this misses the point. Backpressure addresses throughput issues, while Costa's suggestions focus on improving output quality rather than simply reducing velocity.

The lean manufacturing framework offers a better lens. Beyond waste reduction, lean tackles unpredictable human input by designing resilient processes. Traditional management blames workers for mistakes, demanding perfection with threats of consequences. Lean thinking instead accepts that people aren't perfectly reliable and builds systems accordingly.

Three lean practices apply directly to AI-integrated development. Single-piece flow processes one item at a time, enabling downstream rejection before excess defects accumulate. Autonomation lets machines detect anomalies and stop automatically. Poka-yoke designs quality into processes by construction, preventing errors rather than catching them post-facto.

Managers who blame workers for quality problems face an obvious contradiction with robots—they cannot reasonably hold machines accountable. This makes the lean philosophy essential: systems must assume imperfection and build resilience accordingly, whether handling human developers or autonomous code generators. As Deming noted, a bad system beats good people every time.