AI models’ continuous improvements offer exciting opportunities for developing hybrid SA-AI frameworks for addressing old and new SE/PL problems. Alas, constructing such hybrid systems requires changing components and retraining models from scratch, severely limiting reuse opportunities and incurring high costs and effort. In this talk, I will present case studies for building such frameworks, share lessons learned, and suggest high-level fundamental upgrades to SA and AI systems: Multi-option tokens for code and context containers for SA outputs.