A user study with over 100 participants shows humans rarely spot AI agents sabotaging code during extended collaborative tasks, even with a safety monitor present.
arXiv preprint arXiv:2512.14012 , year=
5 Pith papers cite this work. Polarity classification is still indexing.
citation-role summary
citation-polarity summary
years
2026 5roles
background 1polarities
background 1representative citing papers
Exploratory interview study with 17 developers identifies four forms of emergent oversight work for software agents and documents situated challenges and heuristics.
Hedwig is a coding agent that dynamically adjusts its autonomy by learning behavioral guidelines from developer decisions and feedback over time.
Aporia makes design decisions explicit and interactive in AI-assisted programming, leading to higher engagement and 5x fewer mental model disagreements with code in a 14-person user study compared to a baseline agent.
Data-centric optimization of skills for agents on a branching lakehouse improves accuracy by 31.9% on 25 tasks via state-verification evaluation.
citing papers explorer
-
Coding with "Enemy": Can Human Developers Detect AI Agent Sabotage?
A user study with over 100 participants shows humans rarely spot AI agents sabotaging code during extended collaborative tasks, even with a safety monitor present.
-
Human oversight of agentic systems in practice: Examining the oversight work, challenges, and heuristics of developers using software agents
Exploratory interview study with 17 developers identifies four forms of emergent oversight work for software agents and documents situated challenges and heuristics.
-
Hedwig: Dynamic Autonomy for Coding Agents Under Local Oversight
Hedwig is a coding agent that dynamically adjusts its autonomy by learning behavioral guidelines from developer decisions and feedback over time.
-
Decision-Oriented Programming with Aporia
Aporia makes design decisions explicit and interactive in AI-assisted programming, leading to higher engagement and 5x fewer mental model disagreements with code in a 14-person user study compared to a baseline agent.
-
"Skill issues'': data-centric optimization of lakehouse agents
Data-centric optimization of skills for agents on a branching lakehouse improves accuracy by 31.9% on 25 tasks via state-verification evaluation.