I propose a relational framework that delineates three fundamental agent roles: (1) assistant, (2) collaborator, and (3) mentor. Each represents a distinct relationship paradigm characterized by different assumptions about agency, expertise, and initiative. This tripartite model provides a structured approach for analyzing agent technologies across domains and applications. By articulating these role distinctions, we can better understand the technical requirements, ethical implications, and design considerations for each category.

Role-1: Assistant

Assistant agents operate under a paradigm where the human user maintains primary agency and decision-making authority. These agents execute delegated tasks without questioning user intent, embodying a “do what I say” relationship model. The underlying assumption is that the human possesses sufficient knowledge to determine appropriate tasks and evaluate outcomes effectively.

Examples of assistant agents include systems for booking travel arrangements, shopping, scheduling meetings, or executing routine data processing tasks. The primary value proposition is efficiency—the agent saves time and effort by handling well-defined tasks that the human could perform but chooses to delegate.

Role-2: Collaborator

Collaborator agents operate under a paradigm of shared agency and complementary expertise. Unlike assistants, collaborators contribute domain knowledge and capabilities that the human may lack, creating a synergistic “let’s work together” relationship. The underlying assumption is that neither party possesses complete knowledge or capabilities, but together they can achieve outcomes superior to what either could accomplish independently.

Examples include co-creative systems in design, coding assistants that suggest implementations, diagnostic systems in healthcare, or scientific discovery tools that identify patterns in complex datasets. The primary value proposition is augmentation—enhancing human capabilities through complementary expertise and perspective.

Role-3: Mentor

Mentor agents operate under a paradigm where the agent assumes responsibility for user growth and development. These agents proactively identify opportunities for human learning and capability enhancement, embodying a “help me grow” relationship. The underlying assumption is that the agent possesses expertise the human wishes to acquire and can structure experiences to facilitate this development.

Examples include educational systems that adapt to learning patterns, fitness coaches that progressively challenge users, or professional development tools that identify skill gaps and suggest improvement strategies. The primary value proposition is transformation—changing human capabilities and knowledge over time rather than merely complementing or executing tasks.

I’m not suggesting that we have to pigeonhole any agent into just one of these three roles, but thinking through which role(s) an agent can play could help us not just answer that ‘how’ question, but also the ‘why’ question about that agent. Or it may also lead to a realization that an agent is not the right solution. I am an “AI pragmatist”, and so I would like to see AI, however you define it, help us solve our problems in a responsible manner. If agents are doing that, then sure, let’s build and deploy them. But I wouldn’t want to do that without thinking through what value they would really add, what role they could play in a given situation, and what are the implications for individuals and society at large with agents in the picture.