Accenture
Verified evidenceCreator to judgeDevelopers can accept, edit, or reject inline suggestions while build, review, and merge controls remain.
Software engineering · Global
Collections: Human still decides
Executive brief
Real-world productivity evidence is more modest and noisier than lab claims; adoption and quality controls matter.
AI value · Pull requests completed per week
VerifiedAccess was associated with 7.51%-8.69% more pull requests per week, depending on specification.
Preliminary estimates were imprecise and statistically significant only under weighting for periods with stronger uptake differences.
Before
Writes implementation and test code manually. → Reviews submitted code and decides whether to merge.
After
Receives inline code suggestions and accepts, edits, or rejects them. → Reviews the complete human-owned change.
Human boundary
Developers decide whether suggestions enter a change; reviewers and repository controls decide merge.
Why it matters
Developers can accept, edit, or reject inline suggestions while build, review, and merge controls remain.
Before
How the work ran before the change.
Developer
Writes implementation and test code manually.
ControlTests and code review.
Reviewer
Reviews submitted code and decides whether to merge.
ControlStandard engineering governance.
What changed
Developers can accept, edit, or reject inline suggestions while build, review, and merge controls remain.
Decision rightHuman moves from creator to judge
After
How the same work runs now.
Developer with Copilot
Receives inline code suggestions and accepts, edits, or rejects them.
ControlDeveloper review, tests, scanning, and policy.
Reviewer
Reviews the complete human-owned change.
ControlUnchanged merge authority.
Exception path
Developers reject unsuitable output and code manually; failed tests or reviews block the change.
Work removed
Decision authority
Developers decide whether suggestions enter a change; reviewers and repository controls decide merge.
Before
Developer
Writes implementation and test code manually.
Control: Tests and code review.
Reviewer
Reviews submitted code and decides whether to merge.
Control: Standard engineering governance.
After
Developer with Copilot
Receives inline code suggestions and accepts, edits, or rejects them.
Control: Developer review, tests, scanning, and policy.
Reviewer
Reviews the complete human-owned change.
Control: Unchanged merge authority.
Work that left the path
Human role before
Developers authored code directly and reviewers governed merges.
Human role after
Developers supervise suggestions and remain accountable for code; reviewers retain merge control.
AI roleDecision mode: inline generation. Suggests code in context but cannot submit or merge.
Randomized Accenture developers without Copilot access.→Access was associated with 7.51%-8.69% more pull requests per week, depending on specification.
Field experiment beginning July 2022. · 311 Accenture developers in the research sample.
Preliminary estimates were imprecise and statistically significant only under weighting for periods with stronger uptake differences.
Anti-pattern
Using pull-request count alone as proof of engineering value.
Questions
Portability conditions
Reputation risk
medium
Evidence and authority
Current · updated
1 independent; publication outcomes are verified.
Bundle 1.0.0 · reviewed 2026-08-23 · stable ID bc88eb70918fc8a9
Google · Creator to judge
Large migrations can combine machine-found locations, LLM edits, automated checks, and engineer review.
GitHub Copilot research experiment · Creator to judge
Inline generation can compress a bounded task but not prove lifecycle productivity.
Meta (Facebook) · Creator to judge
Routine static-analysis bugs can receive machine-proposed patches behind compilation and analyzer-clean gates.
Anonymous Fortune 500 software company · Creator to judge
AI can diffuse expert practice, with value concentrated among novices.