David's approach: Using multi-agent parallel audits before every major ship

Across multiple sessions, David applied a consistent pattern before shipping significant changes:

launch 6 gpt 5.4 xhigh subagents, feed the plan to them, and have them go audit your work to see if its fully implemented end to end, if itll work in production as expected, and if itll work reliably and stably and if all race conditions, failure modes, or brittleness has been solved for

This appeared in the lecture notes generation popup removal, the workspace parity work, the canvas file metadata heal, and the audio generation feature. The subagents were given the plan document and the implementation, tasked with finding gaps the main agent had missed. David then reviewed audit findings before approving deployment.