AI2030 Responsible AI Tools Marketplace

Key Takeaways

Autonomous coding agents create a "hidden drift" problem: As AI agents work faster and make more decisions independently, there's a growing gap between what they actually do and what developers intended—a problem that traditional code review methods miss because they only look at final outputs, not the agent's reasoning process.
Current oversight tools are insufficient: Standard code comparison tools (diffs) and existing AI explainability methods fail to capture the full picture of how agents behave over time, across multiple tool uses, and throughout a system's architecture, leaving developers with incomplete visibility into autonomous decision-making.
Process transparency matters as much as code quality: The framework prioritizes making visible how agents reach decisions (through monitoring their reasoning and tool choices) rather than just what code they produce, ensuring humans maintain meaningful comprehension of autonomous actions—which the authors argue is essential for responsible oversight.
Different users need different kinds of visibility: Casual users benefit from seeing the structural logic behind agent actions (catching problems that "work" but violate intent), while professional developers need richer context for code review without adding review burden—suggesting responsible AI oversight must be tailored to user expertise levels.
Human oversight requires intentional system design: The paper argues that substantive human control over autonomous agents isn't automatic; it requires deliberately building explainability into agent systems from the start, treating oversight as a core design requirement rather than an afterthought.

Abstract

As autonomous coding agents become deeply embedded in software development workflows, their high operational velocity introduces a critical oversight challenge: the accumulating divergence between agentic actions and architectural intent. We term this process agentic entropy: a systemic drift that traditional code diff-based and HCXAI methods fail to capture, as they address local outputs rather than global agentic behaviour. To close this gap, we propose a process-oriented explainability framework that exposes how agentic decisions unfold across time, tool calls, and architectural boundaries. Built around three pillars (conformity seeding, reasoning monitoring, and a causal graph interface) our approach provides intent-level telemetry that complements, rather than replaces, existing review practices. We demonstrate its relevance across two user profiles: lay users engaged in vibe coding, who gain structural visibility otherwise masked by functional success; and professional developers, who gain richer contextual grounding for code review without increased overhead. By treating cognitive drift as a first-class concern alongside code quality, our framework supports the minimum level of human comprehension required for agentic oversight to remain substantive.

Camera-ready version of the position paper accepted to the Human-Centered Explainable AI (HCXAI) Workshop at CHI 2026

Key Takeaways

Autonomous coding agents create a "hidden drift" problem: As AI agents work faster and make more decisions independently, there's a growing gap between what they actually do and what developers intended—a problem that traditional code review methods miss because they only look at final outputs, not the agent's reasoning process.
Current oversight tools are insufficient: Standard code comparison tools (diffs) and existing AI explainability methods fail to capture the full picture of how agents behave over time, across multiple tool uses, and throughout a system's architecture, leaving developers with incomplete visibility into autonomous decision-making.
Process transparency matters as much as code quality: The framework prioritizes making visible how agents reach decisions (through monitoring their reasoning and tool choices) rather than just what code they produce, ensuring humans maintain meaningful comprehension of autonomous actions—which the authors argue is essential for responsible oversight.
Different users need different kinds of visibility: Casual users benefit from seeing the structural logic behind agent actions (catching problems that "work" but violate intent), while professional developers need richer context for code review without adding review burden—suggesting responsible AI oversight must be tailored to user expertise levels.
Human oversight requires intentional system design: The paper argues that substantive human control over autonomous agents isn't automatic; it requires deliberately building explainability into agent systems from the start, treating oversight as a core design requirement rather than an afterthought.

Abstract

Camera-ready version of the position paper accepted to the Human-Centered Explainable AI (HCXAI) Workshop at CHI 2026

Beyond the 'Diff': Addressing Agentic Entropy in Agentic Software Development

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

Beyond the 'Diff': Addressing Agentic Entropy in Agentic Software Development

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links