Building Persona-Based Agents On Demand: Tailoring Multi-Agent Workflows to User Needs Current multi-agent AI systems often rely on rigid, pre-defined architectures where roles, communication patterns, and interaction flows are hard-coded by developers. This "one-size-fits-all" approach makes it difficult for systems to adapt to the unique needs, preferences, and contexts of individual users. This paper proposes a new paradigm: on-demand persona-based agent generation. Instead of using fixed agent profiles, the system dynamically creates and configures agents at run-time, tailoring their behavior, reasoning, and communication style to match the specific user and the task at hand. Moving Beyond Fixed Architectures Traditional agentic platforms are limited by design-time choices, where developers must anticipate every role and interaction protocol in advance. The authors argue that by treating personas as flexible, generative constructs rather than static templates, platforms can become significantly more adaptable. In this model, the system does not force users to conform to a pre-set structure; instead, the system synthesizes the necessary agents on the fly, allowing for a more natural and personalized collaboration between humans and AI. The Four-Step Generation Pipeline The proposed system uses an orchestrator to manage the creation of agents through a four-step session-based pipeline: Query Analysis: The system captures the user's intent and characteristics, then decomposes the request into a structured plan of discrete, dependent tasks. 2. Agent Generation and Instantiation: The orchestrator crafts specific personas for each task, defining the required roles, competencies, and communication styles. These personas are then used to initialize a pool of specialized agents. 3. Agent Assigning and Execution: The orchestrator manages the workflow, assigning tasks to the appropriate agents and ensuring that information flows correctly between them based on the task dependencies. 4. Answers Aggregation and Displaying: The system collects the results from all agents, resolves any inconsistencies, and merges the outputs into a single, coherent response tailored to the user’s preferred format. Benefits of Dynamic Adaptability By integrating persona-based generation as a core architectural pillar, this approach aims to improve the quality of human-AI interaction. Because the system can adjust its behavior based on the user's expertise level, background, or specific task context, it reduces the friction often associated with complex multi-agent systems. This design ensures that the platform remains accessible and meaningful, as the system evolves its own configuration to align with the user throughout the interaction session. Future Implications for Design The authors suggest that this shift toward runtime adaptability represents a significant change in how agentic platforms are designed. By moving away from rigid hierarchies, developers can create systems that are more responsive to real-world scenarios. This framework positions on-demand persona generation as a key tool for empowering end-users, ensuring that as agentic platforms grow in complexity, they remain flexible enough to provide personalized and contextually appropriate support.