Governing AI-Generated Code: Why Individual Speedups Do Not Become Organizational Outcomes

AI coding assistants have reached near-universal adoption, and controlled studies consistently show large task-level speedups. Yet organization-level telemetry shows those gains frequently fail to convert into improved delivery or business outcomes. This paper examines that gap as an engineering governance problem rather than a tooling problem. Drawing on recent field data — longitudinal code-quality analysis, a randomized controlled trial on skill formation, labor-market evidence, and security benchmarking — we identify three structural mechanisms that absorb AI-assisted speed: a review bottleneck that scales with generation rather than comprehension, comprehension debt that is costlier to recover than conventional undocumented code, and a verification asymmetry in which correctness is cheaply checked while security and efficiency are not. We propose a risk-tiered autonomy framework that calibrates permitted agent autonomy and required human verification to the consequence of a change, and a measurement set for detecting absorption before it manifests as an incident.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-21
DOI
https://doi.org/10.5281/zenodo.24423777
Primary Topic
Software Engineering Research
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Governing AI-Generated Code: Why Individual Speedups Do Not Become Organizational Outcomes

Rakesh Rai
Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research
article

Governing AI-Generated Code: Why Individual Speedups Do Not Become Organizational Outcomes

Rakesh Rai
article en

Abstract

AI coding assistants have reached near-universal adoption, and controlled studies consistently show large task-level speedups. Yet organization-level telemetry shows those gains frequently fail to convert into improved delivery or business outcomes. This paper examines that gap as an engineering governance problem rather than a tooling problem. Drawing on recent field data — longitudinal code-quality analysis, a randomized controlled trial on skill formation, labor-market evidence, and security benchmarking — we identify three structural mechanisms that absorb AI-assisted speed: a review bottleneck that scales with generation rather than comprehension, comprehension debt that is costlier to recover than conventional undocumented code, and a verification asymmetry in which correctness is cheaply checked while security and efficiency are not. We propose a risk-tiered autonomy framework that calibrates permitted agent autonomy and required human verification to the consequence of a change, and a measurement set for detecting absorption before it manifests as an incident.

Zenodo (CERN European Organization for Nuclear Research)
Software Engineering Research
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon