Exception handling has a commonly agreed semantics in many programming languages. When an operation is called with inappropriate conditions, the control flow of the program is reversed back to the caller to trigger some handling mechanisms. In Multi-Agent Systems (MAS), this semantics applies to the code of agents, but it does not cover the kind of exceptions that occur at the agent level. For instance, the usual handling semantics does not address the cases where the plan of an agent fails and re-planning is required. In fact, the agent code does not necessarily encounter a programming fault or a 'catch' clause in such a case, but the agent has yet to deal with an exceptional situation. In this paper, we survey the literature on exception handling with the aim to demonstrate that research needs to be conducted in the case of MAS due to their openness, heterogeneity, and the encapsulation of agents. We identify research directions from the survey, and we present a simulation scenario to illustrate the occurrence of agent-level exceptions in a simple case. The current result of the survey analysis is that agent exceptions mechanisms should rely on the proactivity of agents, on exploiting the agent environment, on collaborative handling mechanisms, and on the potential links between code-level and agent-level exceptions.