Can a feasible plan still need reconsideration?

Yes. Feasibility does not establish that its goal remains the preferred objective.

Does goal monitoring require an agent to act autonomously on every change?

No. It may identify a decision for a human or another authorized process.

Is the highest-scoring goal always acceptable?

No. Constraints, permissions, obligations, and the quality of the scoring assumptions also matter.

AI Glossary / MLOps and deployment / Intermediate

Goal monitoring

Definition

Goal monitoring is the process of checking whether the goals an agent is currently pursuing remain the most appropriate goals to pursue. In automated planning, it asks whether a better goal set is available before continuing with an action. It concerns the choice of objectives, rather than only whether the next action is possible or the existing plan remains feasible.

What it means

An agent may be following a valid plan toward a goal that has become less useful. New requests, deadlines, resource changes, or revised priorities can change which objectives should receive attention.
Goal monitoring distinguishes being able to finish something from having a good reason to keep prioritizing it. A scheduling agent, for example, might still be able to complete a routine task while a newly received, authorized urgent task deserves the available time.
The agent needs rules for comparing goals and boundaries on what it may change. Goal monitoring does not authorize inventing business objectives or disregarding instructions whenever its scoring method prefers something else.

How it works

Represent the current goal set, relevant context, candidate alternatives, and the rules governing goal selection. Observe changes that may affect those rules, such as an updated deadline or an authorized change in priority.
Evaluate whether the current goals remain suitable and whether an alternative is both permitted and preferable. Consider feasibility, expected benefit, effort, switching cost, and obligations that must continue to be met.
If the goal set changes, the existing plan may need repair or replacement. Record the reason and preserve necessary information about interrupted work. In some systems the appropriate result of goal monitoring is a recommendation or request for a decision, rather than an autonomous goal change.

Three practical examples

A scheduling agent can reconsider routine work when an authorized urgent request arrives and the scheduling policy specifies how that priority should be handled.
A maintenance-planning system can review whether its planned inspections still target the most relevant equipment after receiving updated condition information.
An information-gathering assistant can recognize that a clarified user question changes the objective of its research and revise its plan within the new request.

Think of it this way

Think of a gardener deciding which garden bed needs attention today. The tools may work and the original watering route may remain possible, yet new observations can make another bed more pressing. The analogy concerns choosing the objective. It does not grant an automated system discretion beyond the owner's priorities, constraints, or authorization.

Worked example

Consider an invented scheduling agent with four available time units. Its current goal A requires all four units and has an illustrative benefit score of eight. A newly authorized goal B requires three units and has a benefit score of twelve. Assume the policy permits choosing either goal, the tasks cannot both be completed within the available time, and the scores use the same agreed scale.
If neither task has started and all other considerations are equal, the comparison favors B because twelve exceeds eight. The original plan for A may still be perfectly feasible; the reason to reconsider it is the preferred objective, not a failed action.
Now introduce a switching cost of five score units because A has already entered a stage that needs cleanup before interruption. Under this simplified rule, switching yields twelve minus five, or seven net units, which is less than the eight associated with continuing A.
Finally, suppose A is a mandatory commitment that the policy does not permit the agent to abandon. The optional score comparison cannot override that constraint. The agent must preserve A or escalate the conflict according to the applicable process.
These invented values demonstrate the role of priorities, switching costs, and constraints. They are not a general decision formula, and they do not imply that diverse human objectives can always be reduced to one numerical score.

Illustrative case study

In a fictional internal research project, an assistant begins preparing a broad comparison of several software tools. The project owner then clarifies that the immediate decision concerns compatibility with one existing system.
The assistant updates the research objective to answer the clarified question and revises the remaining work accordingly. It retains useful earlier notes but does not keep expanding the broad comparison merely because that was the original plan.
The case illustrates goal monitoring grounded in updated user intent. The assistant's response follows the revised task; it does not independently substitute a different organizational objective.

Common misconceptions and mistakes

Do not confuse goal monitoring with checking whether a goal has already been achieved. Achievement checking is related, but the distinctive question here is whether the current goal set remains preferable.
Do not confuse it with plan monitoring. A plan can still succeed even when its objective is no longer the best permitted priority.
Do not ignore switching costs or commitments. Constantly chasing the latest apparent opportunity can prevent any useful work from finishing.
Do not treat a numerical score as an authorization mechanism. Permissions and hard constraints must remain explicit. Also, avoid changing goals on weak or stale evidence without accounting for uncertainty and the cost of unnecessary interruption.

Business use

Goal monitoring is useful where priorities change while automated work is under way. Organizations can make it practical by defining which goals may be reconsidered, which constraints remain fixed, and which changes require a responsible person's decision.
Useful records explain what changed, which alternatives were considered, and why work continued or shifted. The business purpose is to keep effort aligned with current authorized objectives while avoiding both blind persistence and unstable switching. Its design therefore involves workflow policy as well as planning algorithms.

Check your understanding

An agent's current plan remains feasible, but an authorized priority update makes another objective more important. Which monitoring level addresses that change?

Answer: Goal monitoring, because the issue concerns which objective to pursue rather than whether the existing actions remain possible.

Related terms

Execution monitoring
Plan monitoring
Action monitoring
Utility
goal selection
replanning
priority
and constraint

Learn next

Describe a workflow with two competing goals. State the allowed priority rule, a switching cost, and one obligation that cannot be overridden. Explain which changes the system may make itself and which require an authorized decision.

Sources

Sources: Russell and Norvig, Artificial Intelligence: A Modern Approach, fourth-edition teaching slides, Chapter 11, slide 19, university-hosted publisher material: https://zoo.cs.yale.edu/classes/cs370/lectures/slides/chapter11_4e.pptx.pdf

Previous: Goal formulation Next: Goal test