To Copilot and Beyond: 22 AI Systems Developers Want Built

A new research paper from Microsoft and academic collaborators reveals a significant gap in current AI developer tooling. While developers spend roughly 10% of their workday writing code, most AI assistance—like GitHub Copilot—targets just that fraction. The study, "To Copilot and Beyond: 22 AI Systems Developers Want Built," surveyed 860 Microsoft developers using a human-in-the-loop, multi-model council-based thematic analysis to identify where they actually want AI support across their entire workflow.

The research identifies 22 desired AI systems across five task categories, pointing to a growing "right-shift burden" where developers need quality signals earlier in their workflow to keep pace with accelerating code generation. Developers expressed clear boundaries, wanting AI to absorb the "assembly work" surrounding their craft—never the craft itself. This pattern, termed "bounded delegation," reveals that professional identity dictates where AI should stop, suggesting tooling value lies as much in precise limitations as in capabilities.

Key constraints developers placed on these systems include explicit authority scoping, provenance tracking, uncertainty signaling, and least-privilege access throughout the development process. The findings suggest future AI tools must balance automation with developer control, embedding quality assurance earlier while respecting professional boundaries that developers fiercely protect.