| Inversion of control | Design to provide greater control from service provider to consumer, thus creating a paradigm shift from service personalization to personal AI-agents that increase perceived control to vulnerable consumers, thereby providing greater agency across the breadth or depth vulnerability | Provides more power to the consumer. Reduces goal divergence from a consumer perspective as their CX shifts from passive to active control and decision making as they formulate their cognitive, emotional, behavioral, sensory and social responses | Project VRM [IV] - Vendor Relationship Management, where the idea is to shift control from a Customer Relationship Management (CRM) operated by a providers to consumers who manage and control the relationship with vendors |
| Delegated authority and decision-making alignment | Design attribute that allows consumers to grant entitlements to AI-agents to make decisions on their behalf, thereby potentially reducing the cognitive and emotional burden of vulnerable consumers. The designs should allow for specific actions to be performed by AI-agents that are delegated or not | Could reduce cognitive, emotional, sensorial, behavioral and/or social load for the consumer, thus improving well-being | Algorithmic trading AI [V]-agents such as the one provided by ig.com in the UK are designed with provisions to gain delegated authority to be able to trade in the market on behalf of consumers |
| Adapts to vulnerability | This attribute entails abilities within personal AI-agents to natively adapt to the vulnerability context of consumers. Detection of breadth and depth of vulnerability becomes a key design decision that can ensure an effective implementation of AI-agents that are adaptable | By dynamically adapting to the breadth and depth of a consumers vulnerability in a particular context, the AI-agent can customize compensation for specific accessibility limitations of the individual in context facilitating improved or magnified cognitive, emotional, behavioral, sensory and/or social responses. This compensatory adaptation should empower the consumer and improve their well-being via achievement of all exchange goals | Neuralink Blindsight (Moeed, 2024) illustrates a piloted implementation of an AI-agent designed to adapt to the cognitive vulnerability of a consumer to help visually impaired consumers |
| Does not exploit vulnerability | This design attribute explicitly specifies personal AI-agents to not exploit the vulnerability of their consumers, instead of this being an implicit understanding in good faith. Safeguarding and prevention of misuse should be part of the AI-agent designs efforts upfront rather than an afterthought | AI-agent prevented from exploiting information asymmetry ensures equitable value transfer, safety and dignity of the principal (consumer) minimizing risk of negative cognitive, emotional, sensorial, behavioral and/or social responses. Thus, this should help prevent economic, physical or psychological harm and/or reduced well-being | Kwaai.ai [VI] provides an example where consumers retain their personal and behavioral data, thereby having greater perceived control in their service interactions |
| Interoperability | This attribute specifies AI-agents to have the ability to interchange data and services with other AI-agents. Integrating with the processes of service providers as well as adoption of standards seamlessly can increase interoperability | Reduces the risk of consumer being trapped into a specific technological platform and experiencing goal divergence that they cannot mitigate via exit, leading to negative cognitive, emotional, sensorial, behavioral and/or social responses through failure to realize all exchange goals | Solid specification on the W3C web storage [XVII] provides an example of standards that could be used by personal AI-agents to improve interoperability |