What is Agency Decay

AI is such a versatile technology that it is easy to be tempted to be pulled into using it indiscriminately. Dr Walther flags 4 pervasive risks to watch out for as we grow comfortable using AI in our day-to-day work. She categorizes these risks under a larger problem that she calls “agency decay.”

Dr Walther defines agency decay as a “gradual erosion of a person’s ability and willingness to observe carefully, think independently, choose deliberately, and act responsibly.” This happens when we get so comfortable with the convenience of using conversational AI as a thinking partner that it becomes a default setting. Then without knowing it, our own ability to think independently gets reduced and agency decay sets in

4 Stages of Agency Decay to Watch Out for

Agency decay or erosion in independent cognition, doesn’t happen overnight. It has the following 4 stages, says Dr Walther

• Experimenting – This is the “playful and practical” stage. The LLMs help new employees sound clearer. A non-native speaker uses them to write more fluently. Or a busy sales professional might rely on LLMs to quickly create client meeting notes.

  At this stage, the human remains in charge. But there is a subtle risk, Dr Walther points out. Productivity improvements may encourage us to assume that faster output means better work. The assumption becomes risky in situations that demand “judgement, context, ethics or accountability.” Continued reliance on AI tools could also make it difficult to work without them even for simple tasks

• Integrating – In this phase, use of LLMs and generative AI becomes part of the workflow. “This,” observes Dr Walther, “is where the relationship changes.” Once we start viewing Language Models not just as tools but as collaborators, there is an ever-present temptation to incorporate their output directly into our drafts, reports and recommendations.

  Take the case of salespersons prospecting a potential buyer from new industries or businesses they are not familiar with. They may use an AI tool to learn more about the new industry or business. The research output they receive from the tool often sounds so convincing and authoritative that they may feel like using it as it is in building a proposal. As familiarity grows, trust in AI rises. This is where cognitive offloading transitions into belief offloading. AI output starts getting accepted as correct even when evidence doesn’t seem to support it. The question slowly shifts from “Is this task suitable for AI?” to “What did the AI say?”

• Reliance – In the next stage AI becomes the default starting point for everything. Work gets done by AI tools with the output being reviewed by people. But this marks a shift from “human thought to machine generation.” Usage, however extensive, is not the problem. Risk is in relying on outputs without examining and validating them independently.

  Dr Walther advises – “The lesson is simple and uncomfortable: The more people trust the tool, the more deliberately they must protect the habit of questioning it.” In a sales context, accepting whatever AI offers without questioning, could result in costly and time-consuming rework and could create a poor impression with clients

• Depending – When individuals or teams struggle to perform, decide, or verify without AI, it signals entry into the stage of dependence. The negative effects at this stage are not limited only to the individual. They affect teams and organisations also.

  Dr Walther raises a disturbing question – “What happens to skill formation when the first draft is increasingly outsourced?”

4 As Framework to Retain Human Edge

• Awareness – Begin by noticing what you are doing. Before using AI, ask: What exactly am I handing over — Drafting, reasoning, judgment, empathy, verification, accountability?
• Appreciation – Appreciate work situations where AI genuinely expands human capacity. Entrust tasks like summarizing, translating, comparing, simulating, or picking up blind spots in which AI can augment and speed up human effort. Retain higher order tasks like problem framing, values setting, context appreciation, ethics and quality standards with humans
• Acceptance – Accept that AI has its limits even when it sounds confident and sure. Also recognise that making people more productive with AI could also make them less skilled. People should have sufficient opportunities to practice the underlying skills without automation. For instance, salespeople could decide to continue writing some percentage of mails by themselves without relying on AI support to ensure retention of mail writing skills
• Accountability – Regardless of who generates the output, AI or individuals, accountability rests with humans in the system. So, cultivating a “human first draft” habit for important thinking is a good practice. This also includes mandatory source checks, disclosures of AI usage and assigning human owners for AI-supported decisions

Winners of the future will be those organisations that treat human agency, autonomy, judgment and trust, as business assets to be preserved and grown in tandem with the usage of AI

The Knowledge@Wharton article Is Overreliance on AI Causing Agency Decay? by Dr Cornelia C Walther can be read here