In April 2019, the Event Horizon Telescope (EHT) released the first-ever image of a black hole. To produce this image required days to weeks using high-performance and distributed computing. In the next decade, the next-generation EHT will come online, increasing the complexity of black hole imaging by 2-3 orders of magnitude, thus making current tools impractical. Comrade, a Julia-based Bayesian black hole imaging code, has recently been developed to enable imaging of the next-generation EHT data. By utilizing Enzyme and Julia, we achieved 10-100 times higher performance compared to current tools. This talk will overview the problem of black hole imaging and our ongoing experiences of incorporating Enzyme into Comrade. Additionally, we will highlight how Enzyme solved our scaling problems compared to other autodiff tools like Zygote. Finally, we will discuss what challenges remain and the features we need to use Enzyme in our entire code base moving forward.