For decades, Denard Scaling produced smaller, faster transistors without increasing power density. Faster transistors led to faster processors that sparked a virtuous cycle of software development. However, Denard Scaling will soon reach its limits. At the ReROUT Lab at The Ohio State University, we are studying models for software design and management in the post-Denard era. Our work targets Internet services where response time affects profit. Service level objectives are widely used to align response time and business costs. Zoolander is a model-driven middleware for key-value stores that parallelizes lookups cost effectively, trading costly service-level violations for less costly replication. When systems-level management, e.g., replication, does not suffice to meet SLOs, approximation may be an answer. Ubora measures the effect of approximation on the quality of answers produced by a service. Consider the OpenEphyra question-answering system, a mature query execution may yield the same answer as a partial execution because of redundancy within the data. We've used OpenEphyra in Jeopardy competitions with 7th and 8th graders, as a part of our outreach efforts. We used Ubora to control approximation during the competitions. Early outcomes suggest that approximation can significantly reduce response time without degrading answer quality too much (i.e., our system still wins!).