The robustness of smart grids is challenged by unpredictable power peaks or temporal demand oscillations that can cause blackouts and increase supply costs. Planning of demand can mitigate these effects and increase robustness. However, the impact on consumers in regards to the discomfort they experience as a result of improving robustness is usually neglected. This paper introduces a decentralized agent-based approach that quantifies and manages the tradeoff between robustness and discomfort under demand planning. Eight selection functions of plans are experimentally evaluated using real data from two operational smart grids. These functions can provide different quality of service levels for demand-side energy self-management that capture both robustness and discomfort criteria.