Skip to content

Antagonistic AI design agent with steerable stakeholder pushback in Miro

Design · All designs
Evidence · 2 claims (2 for) · 1 study (1 causal), q3 · 0 of 1 report an effect size · 2 claims rest on one study

Description

An LLM-powered agent built in Miro (Next.js, Miro SDK, GPT-4.1, Whisper, GPT-4o) that enacts constructive conflict for interaction design students. In the Agent Intervention Space, "the agent generates a first round of four pushback points anchored in the designer's current proposal; the designer interacts with those points to steer the design direction; the agent then generates a second round informed by that steering." Each pushback point pairs a stakeholder perspective with a specific challenge. Designers steer via point-level tagging (useful, not useful, custom) and stance-level Consensus and Command frames. A sub-agent retrieves a database of prior constructive-conflict examples to ground critiques.

Design Implications

Context

Requirements

  • Multimodal context gathering (text, visuals, think-aloud) so pushback is anchored in the designer's contextual exploration
  • Steering mechanisms (tagging and Consensus/Command frames) so designers retain agency while engaging with critique

Constraints

  • The agent could not infer relationships between linked ideas from think-aloud input
  • It cannot show how common, important, or grounded synthetic concerns are in lived experience and does not substitute for engagement with the real public
  • A few Miro actions not exposed by the open API were handled by the researcher via a predefined, bias-avoiding protocol

Target Learners

  • Novice interaction designers (design students trained in interaction design but novices at navigating stakeholder conflict)

Learning Goals

  • Engaging constructively with multi-stakeholder tension during early design ideation
  • Reconsidering and revising design decisions in light of stakeholder pushback

Claims

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
Page metadata
Field Value
type design
id antagonistic-design-agent-steerable-pushback
title Antagonistic AI design agent with steerable stakeholder pushback in Miro
description An LLM-powered agent built in Miro (Next.js, Miro SDK, GPT-4.1, Whisper, GPT-4o) that enacts constructive conflict for interaction design students.
status review
generated by process:settle-candidates on 2026-10-10
trust unverified — no one has confirmed this page's content
sources 1 — howard-ziyu-han-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.