The misaligned incentives behind AI coding agents
The industry is shifting focus from training bigger models to optimizing cost and speed. Key techniques include the 'sidekick' agent architecture, which uses parallel processing of multiple LLMs for better price performance.
Read article