Break It 'Til You Make It: Building the Self-Improving Stack for AI Agents