Hybrid Agentic Framework

• Hybrid Agentic Framework is a structured model delineating the distribution of roles, authority, and accountability between humans and AI during task execution.
• It utilizes five orthogonal axes—cognitive complexity, collaboration, creative agency, responsibility, and involvement—to formalize task partitioning and oversight.
• The framework enhances multimodal communication of designer intent, enabling precise negotiation of creative and operational responsibilities.

A hybrid agentic framework defines the structured division and dynamic negotiation of roles, authority, and accountability between humans and AI agents within collaborative workflows. It provides multidimensional axes for specifying how tasks are partitioned, how human intent is communicated, and where creative agency and operational responsibility reside. In design workflows, this framework separates AI’s assistive and autonomous functions from cosupervised, creative, and authority-driven actions of human designers, thereby supporting finely granular negotiation of control and oversight (Wadinambiarachchi et al., 25 Sep 2025).

1. Multiaxial Framework: Five Dimensions of Human–AI Agency

The framework deploys five orthogonal axes to systematically chart and operationalise who acts, how, and with what level of autonomy or responsibility during design tasks:

• Cognitive Complexity:
  • Mundane: mechanical/repetitive (e.g., file renaming, batch resizing)
  • Analytical: pattern recognition, clustering (e.g., meeting summarisation, asset categorisation)
  • Creative: generative, reframing, original synthesis (e.g., generating novel directions)
• Degree of Collaboration:
  • Individual: work executed in isolation by either human or agent
  • Coordinating: interdependent, sequential hand-off (e.g., agent drafts, human revises)
  • Collaborative: parallel, iterative idea exchange
• Creative Agency:
  • Low: assistive, no generative authorship
  • Medium: both parties suggest/reshape, shared authorship
  • High: one (typically human) drives generative direction, agent follows
• Responsibility (Accountability):
  • Minimal: little or no oversight
  • Moderate: shared accountability
  • High: full creative outcome accountability (typically human)
• Involvement (Engagement Level):
  • Passive: agent acts only if summoned
  • Neutral: agent ready, waits for explicit signal
  • Active: agent initiates, monitors, and suggests proactively

Any collaborative scenario is mapped by scoring both human and agent along these axes, yielding a clear, multidimensional “collaboration space.”

2. Authority, Accountability, and Task Positioning

Every workflow can be positioned within this radial five-axis model. For example, a routine file-sorting task configures the agent at “mundane complexity,” “coordinating,” “low agency,” “minimal responsibility,” “passive” and the human as “neutral.” Comparing positions across axes reveals system autonomy, oversight hand-offs, and which party is final arbiter. This supports explicit vocabulary for negotiating boundaries (“I want low agency, full responsibility, and passive involvement from the agent on this curation task”).

3. Intent Communication Beyond Text Prompts

Designer intent frequently exceeds the expressive capacity of textual prompts. The framework incorporates multimodal interaction patterns:

• Guided Prompt Scaffolding: Interactive interfaces walk users through intent specifications, posing follow-up questions.
• Annotations & Tags: Designers annotate wireframes/mood boards via highlights and notes. Standardised semantic tags (e.g., “bold,” “organic,” “outdoors”) add contextual richness.
• Sketch-Based Input: Freehand sketches serve as seeds for generative agent routines; the agent extrapolates variations.
• Voice Commands & Notes: Spoken rationales (e.g., “show me more asymmetrical icons in a muted palette”) are transcribed and acted upon.
• UI Metaphors: Standard menu controls, infinite canvases, spatial audio, and AR/VR hand gestures guide agent actions.

These modalities enable intent explanation that aligns with visual, spatial and operational context, thus supporting richer agentic collaboration.

4. Framework Diagram and Operationalisation

The radial diagram visualises axes with ordered graduations (center-to-edge for mundane→creative, individual→collaborative, etc.). Both human and agent roles are plotted, their interplay defining “collaboration space.” This representation is conceptual rather than formal mathematical; the framework is a planning and negotiation tool for shared authority and accountability.

5. Scenario-Based Configurations

Illustrative mappings clarify real-world authority configurations:

Task	Human Agency / Responsibility	Agent Axis Placement	Notes
Gathering Client Requirements	High / Active	Analytical / Collaborative	AI records, summarises; human validates
Organising Inspiration Assets	Passive / Mundane	Mundane / Coordinating / High	AI tags, groups; agent active, human oversight minimal
Synthesising New Ideas	High / Collaborative / Active	Creative / Collaborative / Medium	AI reframes, proposes; human is principal driver

Such scenarios help teams design agentic platforms tailored to specific tasks, delegating repetitive work to passive automation, or co-opting agents as creative collaborators with accountability mapping.

6. Generalisation, Guidance, and Future Directions

The framework generalises to domains mixing routine, analytical, and creative tasks—e.g., scientific writing, campaign planning, architectural design. It aids tool builders in partitioning control: when to automate, consult or collaborate. Explicit plotting of responsibility and agency supports risk management, especially in scenarios of high-stakes error. User empowerment is prioritised by exposing agent autonomy as a user-controlled parameter (e.g., “autonomy slider”) rather than hidden defaults.

Empirical validation remains a future direction: measuring effects on creativity, trust, efficiency, and supporting dynamic replotting where authority and agency adapt fluidly as tasks evolve.

7. Significance and Conceptual Contribution

By combining five granular axes, the hybrid agentic framework provides both a precise language and schematic for deploying agentic AI in design workflows. It supports explicit and negotiated assignment of control, responsibility, and creativity between humans and autonomous systems. This multidimensional conceptual foundation extends beyond traditional prompt-based generation, enabling robust human–AI partnership and informing the design of next-generation collaborative tools and platforms (Wadinambiarachchi et al., 25 Sep 2025).