TLDR: This research explores how professional designers envision integrating agentic AI into their workflows for inspiration and ideation. Findings show designers want AI to act as coordinators, resource managers, guardians, reframers, and creative catalysts for mundane and analytical tasks, but they wish to retain ultimate creative control. The study also highlights the need for multimodal communication beyond text prompts, such as annotations, sketches, and voice commands, and proposes a framework for orchestrating human-AI collaboration based on task complexity, collaboration degree, creative agency, responsibility, and involvement.
As artificial intelligence continues to evolve, a new concept known as agentic AI is emerging, promising to transform various professional fields, including design. While many are now familiar with generative AI (GenAI) systems that produce content based on prompts, agentic AI takes this a step further by autonomously performing complex tasks with minimal human intervention. This shift has sparked curiosity about how designers envision integrating such powerful AI systems into their creative workflows.
A recent study titled “Imagining Design Workflows in Agentic AI Futures” explores this very question, delving into designers’ perspectives on collaborating with agentic AI. The research, conducted by Samangi Wadinambiarachchi, Jenny Waycott, Yvonne Rogers, and Greg Wadley, investigates how professional designers feel about these advanced AI systems supporting their work, particularly in the crucial early stages of design: inspiration and ideation. You can find the full paper at https://arxiv.org/pdf/2509.20731.
Understanding Agentic AI in Design
The paper highlights that traditional design processes often face challenges like articulating abstract concepts, managing inspiration sources, and avoiding creative fixation. Generative AI tools have already begun to address some of these by assisting with image generation, brainstorming, and creating design variations. However, agentic AI systems are designed to be goal-oriented, adapting to changing circumstances and performing activities with limited direct human supervision. This allows for co-creation, personalized suggestions, and in-context learning, moving beyond the limitations of trained datasets.
The study aimed to understand how designers want AI agents to support and enhance the design process, identifying both opportunities and challenges for future AI-powered creativity support tools (A-AICSTs).
The Study: Imagining a Future with Idy
To gather insights, the researchers employed a “design fiction” study with 10 professional designers. Participants were immersed in a near-future scenario where they collaborated with an AI agent named “Idy” within a design platform called “I-space.” This platform integrated both generative and agentic AI features. The study used a novel “Flip-Flap story card” method, where designers imagined and discussed how Idy could help organize inspiration and ideate for a hiking app project. This approach encouraged participants to tell stories and critique potential AI solutions, revealing their expectations and concerns.
Key Insights from Designers
The findings revealed three overarching themes:
Potential Roles for AI Agents: Designers envisioned AI agents playing several distinct roles:
- Work Coordinator: An AI agent could act as a moderator, creating meeting summaries, task lists, and action points to keep projects on track and manage operational blockers like time constraints.
- Resource Steward: AI could automate mundane tasks such as renaming, organizing, and categorizing design assets based on visual similarities or metadata, making it easier to find and manage files. Interestingly, some designers suggested AI could even purposefully introduce “wrong” information to encourage caution and prevent over-reliance.
- Guardian: Agents could maintain “contextual awareness,” monitoring ideas, auto-saving content, tracking browsing histories, and reminding designers of previous explorations. This role would also involve learning a designer’s style and preferences over time to offer personalized suggestions.
- Reframer: AI agents could nudge designers to recall missed ideas, suggest new trajectories, and even provide counter-examples or alternative viewpoints to prompt exploration of diverse solutions.
- Creative Catalyst: Agents could generate initial ideas as “jumping off points” and offer dynamic “view modes” with varying fidelities (low-fidelity for early exploration, high-fidelity for final decisions) and styles to broaden inspiration and expand conceptual space.
Distribution of Authority: A significant finding was designers’ desire to retain ultimate creative control. They preferred AI to handle routine, low-risk, and mundane tasks, freeing them to focus on creative decision-making. Concerns were raised about AI’s ability to capture nuanced human experiences, empathy, and cultural sensitivity. The level of autonomy granted to AI was seen as dependent on the design stage and task criticality, with more openness to AI suggestions in early exploratory phases and a preference for more control in later stages.
Designers also expressed caution about clients having direct access to AI-powered creativity platforms, preferring clients to provide feedback on presented designs rather than directly manipulating the creative process. They suggested separate client-facing interfaces where AI could mediate suggestions.
Effective Communication with AI Agents: Designers highlighted the limitations of text prompting as the primary input method for AI. They are visual thinkers and found it challenging to translate their imagination into words. They proposed alternative multimodal inputs:
- Enhanced Text Prompting: AI could guide users to structure prompts, provide taxonomies, or offer pre-filled text to assist ideation.
- Annotations and Tagging: Adding explanatory notes or comments to visual content, or using standard tags, could provide contextual information and help AI understand user intent.
- Sketching: Designers suggested using sketches as a visual starting point for AI to generate more refined versions.
- Voice Commands: Voice notes and commands were seen as seamless ways to share intent, with AI transcribing and annotating content in real-time.
Participants also envisioned familiar interfaces, such as menu-driven, canvas-based, smart space, and even augmented/virtual reality (AR/VR) interfaces, for interacting with AI.
A Framework for Human-AI Collaboration
The study synthesizes these findings into a conceptual framework for orchestrating AI agents and human designers in design workflows. This framework defines five dimensions to characterize human-AI interaction:
- Cognitive Complexity: From mundane to analytical to creative tasks.
- Degree of Collaboration: From individual work to coordinated or simultaneous collaborative efforts.
- Creative Agency: The level of creative control and influence over the outcome.
- Responsibility: The accountability for decisions and results.
- Involvement: The engagement level, from passive to active participation.
This framework serves as a decision-support tool, guiding how humans and AI should interact based on task complexity and desired involvement, ensuring that AI empowers designers without diminishing their creative authority.
Also Read:
- UserRL: A Framework for Developing AI Agents That Truly Understand and Assist People
- Researchers’ Digital Brains: Strategies for Knowledge Management in Obsidian
Conclusion
The research underscores that responsible integration of agentic AI into design workflows requires careful consideration of roles, authority distribution, and communication methods. Designers want AI to be adaptable and flexible, taking on active or passive roles depending on the task’s cognitive complexity. Ultimately, AI agents should support designers in ideation and ease mundane tasks, but core creative decisions should remain firmly in human hands. This balanced approach will allow designers to leverage AI’s power while preserving their essential creative agency.
