Steer an antagonistic agent with point-level tagging and stance-level framing¶

Strategy · All strategies
Evidence · 1 claim (1 for) · 1 study (1 causal), q3 · 0 of 1 report an effect size · 1 claim rests on one study

Description¶

The article recommends giving designers explicit mechanisms to steer agent pushback: "(1)point-level tagging, in which they label each pushback point as useful, not useful, or with a custom tag, and discard irrelevant ones; and (2)stance-level framing, in which two optional frames declare the designer's position, namely a Consensus frame for decisions to consolidate and a Command frame for areas inviting additional pressure." These mechanisms let designers stabilize emerging directions while deliberately exposing them to further critique and articulate trade-offs. In the study, 14 of 15 participants used the frames.

Design Implications¶

Context¶

Requirements¶

• An agent that generates pushback points anchored in the designer's current proposal across successive rounds

Constraints¶

• The article tested only one adversarial AI tone and prompt style; future work could explore forms of antagonism from more combative to more supportive

Target Learners¶

• Novice interaction designers working with AI critique agents

Target Learning Goals¶

• Retaining designer agency while engaging with critique structurally rather than as a stream of suggestions
• Making trade-offs explicit by naming which values to prioritize and which to defer

Affordances¶

• Constructive Conflict Agonistic Pluralism Framework

Claims¶

• The antagonistic agent surfaced a broader set of stakeholder concerns than novice designers reached on their own [+W]

Related Strategies¶

Examples¶

-

Key Sources¶

• Howard Ziyu Han, Nikolas Martelaro. (2026). Enacting Constructive Conflicts with AI Agents to Enhance Reconsideration among Novice Interaction Designers. https://arxiv.org/abs/2608.04166