Thoughts on slowing the fuck down

Mario Zechner (creator of the Pi agent framework) argues that orchestrated AI agents remove the human bottleneck, allowing compounding mistakes to accumulate into unmanageable "cognitive debt" before developers notice. Simon Willison amplifies the concern: agent speed collapses weeks of deliberation into hours, causing codebases to evolve faster than our ability to reason about them. Both recommend intentional slowdown — reviewing and understanding agent-generated changes rather than just accepting velocity as a virtue.