Supervisory engineering work is a category of work proposed by Annie Vella and Kelly Blincoe in The Impact of AI Coding Assistants on Software Engineering: A Longitudinal Study to describe the effort software engineers spend directing AI coding assistants, evaluating what they produce and correcting their errors. The authors present it as a new, distinct category that traditional software development lifecycle (SDLC) taxonomies — designing, coding, testing, reviewing — do not capture, and as a possible explanation of where perceived time savings from traditional tasks are being reallocated.
Usage
The construct emerged in that study from convergent quantitative and qualitative patterns. Engineers reported spending less time on most development tasks, with writing code showing the steepest decline, yet verification tasks such as testing and code review rose only modestly, which raised the question of where the effort was going. In open-ended responses participants described shifting from producing code to supervising AI-generated output — directing the assistant, evaluating its suggestions and deciding what to accept, modify or discard — work that does not map cleanly onto existing task categories. The authors suggest that engineers may not recognise this verification of AI output as “testing” or “code review” in the traditional sense.
They break it into three components:
- Directing — specifying intent, crafting prompts, providing context, and iterating when output misses the mark.
- Evaluating — reading AI output and deciding what to accept, modify or reject.
- Correcting — fixing errors, integrating output into existing code, and maintaining consistency.
How much supervision an engineer performs depends, in the authors’ account, on trust calibration: engineers vary in how readily they accept AI output, with some restricting AI to low-risk contexts. They also link the construct to their Productivity-Experience Paradox: the cycle of prompting, reviewing, accepting or rejecting and iterating replaces sustained focus with a rhythm of directing, waiting and evaluating, which they suggest may help explain why flow state eroded.
The concept is at the proposal stage. It rests on one longitudinal survey study, and its authors list as future work testing whether supervisory engineering work is empirically distinct from existing activities, what skills it requires, and whether engineers find it satisfying; if it proves real and distinct, they suggest SDLC models may need updating.