Foundation Models Are Reshaping How Developers Code Together

The study explores how developers use foundation model–powered tools like ChatGPT during open-source collaboration, revealing that shared conversations can enhance collective innovation. Findings highlight gaps in current AI benchmarks, showing that nearly half of code generation prompts contain partial code and many involve multi-turn dialogues. These insights inform better benchmark design, improved prompt-engineering strategies, and the creation of FM tools tailored to diverse developer roles and real-world workflows.