Noisy and ambiguous environments require using the arising uncertainty strategically for efficient behavior. Recent behavioral evidence confirms such strategic use during spatial navigation. Despite this, leading neural models for navigation ignore such uncertainty. I will focus on two examples for revising these models. First, I will show that common ring attractor architectures are able to track uncertainty in heading direction only if they are tuned to feature slow dynamics. Second, I will show that approximate computations that only partially track uncertainty are sufficient to perform close-to-optimal path integration. Overall, these examples highlight the need to account for our strategic use of uncertainty to fully understand how our brains implement spatial navigation.