The satisficing problem for an autonomous agent seeks to minimize the expected search effort. Etzioni considers the case where the agent is constrained by its resources. In this case, the agent must balance the value of a goal against the cost of reaching it. An example of this kind of problem arises where the agent has a number of goals that it can realize and wishes to achieve as many as possible, subject to constraints based on deliberation costs, execution costs, and goal values. The paper presents a model and an architecture based on choosing the action whose marginal expected utility is maximal. By embedding this heuristic in an architecture that keeps a record of previous results, the agent is able to refine its estimates of marginal utility based upon experience. The paper provides a clearly written account of important considerations in problem solving. It would have been worthwhile to include perception as well as deliberation in the analysis of total costs.