Goals, Policies, and Planning
Planning Fundamentals
Goal Representation
A goal is the target state the agent is trying to achieve. In practice, goals should be specific enough to evaluate. For example, "improve customer retention" is less actionable than "identify the top three causes of churn in the last quarter and propose interventions."
Policies and Plans
A policy maps situations to actions, while a plan is an explicit sequence of steps. An agent may use a policy for quick decisions and a plan for longer tasks. Good systems can switch between the two depending on uncertainty and time pressure.
Planning Terms
| Term | Meaning |
|---|---|
| Goal | Desired outcome |
| Subgoal | Intermediate outcome that helps achieve the main goal |
| Plan | Ordered sequence of actions |
| Policy | Rule for selecting actions based on state |
| Constraint | Rule that limits invalid actions |
Planning Under Uncertainty
Agents usually do not know the full state of the world. Planning therefore often means choosing the next best action using partial evidence, not computing an optimal full future.
Which statement best distinguishes a policy from a plan?
Policies are decision rules; plans are step-by-step action sequences.
Correct answer: A policy maps situations to actions; a plan is a sequence of steps
Why should goals be made specific in agent design?
Ambiguous goals make planning and success measurement harder.
Correct answer: Because specific goals are easier to evaluate and break into actions.
A Simple Planning Workflow
-
1
Step 1: Translate the user request into a measurable target.
-
2
Step 2: Break the target into subgoals.
-
3
Step 3: Choose an action for the next subgoal.
-
4
Step 4: Execute and inspect feedback.
-
5
Step 5: Revise the plan if progress is insufficient.
Why is revising a plan important in agentic AI?
Agents operate in changing or partially known environments.
Correct answer: Because new observations can invalidate earlier assumptions