The agency problem

Support has a shadow side. The same person who wants to help can accidentally take over. The same system designed to protect someone can narrow their choices. The same expert language that explains a need can start to define a person by what others think is safest, easiest, or most manageable. In support work, good intentions are not enough. The ethical question is always whether the person is becoming more able to direct their own life, or less.

This is the agency problem. People need help understanding options, communicating preferences, weighing consequences, and recovering from overwhelm. But if support consistently removes the hard parts of choice, the person may be protected from discomfort while also being denied practice in self-direction. The risk is subtle because it often looks caring from the outside.

Supported decision-making offers a better frame. The supporter does not replace the decision-maker. They help the person access information, interpret options, express preferences, and make choices with the right scaffolding. The aim is not independence as abandonment, and not protection as control. The aim is autonomy with support: enough context and relationship around the person that their agency can actually function.

This matters especially in AI-mediated systems. A platform that gives supporters more visibility can easily become surveillance. A platform that suggests next steps can become a command system. A platform that predicts needs can become a quiet override of the person's own voice. If Phil is going to strengthen support networks, it has to keep the person at the centre of the decision loop.

That means designing for context for autonomy, not control. The system should help supporters ask better questions, offer clearer options, notice preferences over time, and step back when the person can act. It should make the learner's own words, interests, choices, and evidence visible, not bury them under expert interpretation. It should help the network support without becoming the expert and without making the person a passive recipient of care.

The best support does not make someone easier to manage. It makes them more able to participate. It builds the capacity to choose, practise, repair, explain, refuse, ask, and try again. In a world where AI can optimise almost anything, preserving agency has to become a design requirement, not a reassuring sentence in the footer.