No Vibes Allowed - Solving Hard Problems in Complex Codebases

Learn advanced context engineering techniques for making AI coding tools effective in large, complex production codebases through this conference talk from AI Engineer 2025. Discover how to overcome the common limitations of AI coding agents that typically struggle with established codebases containing hundreds of thousands of lines of code, often making developers less productive despite their effectiveness on smaller projects. Explore the concept of "frequent intentional compaction" - a family of techniques for strategically structuring context delivery to AI systems throughout the development process. Examine real-world results including successful handling of 300k+ line Rust codebases, accelerating week-long development cycles to single days while maintaining code quality that passes expert review. Gain insights into practical workflows that move beyond typical productivity claims to deliver measurable improvements in complex software environments. Review lessons learned from months of experimentation with current AI models and receive educated predictions about the evolution of AI-assisted software engineering over the next 6-12 months.