Several companies are adopting Kubernetes (k8s) to orchestrate their data and ML pipelines and applications because it enables scalability and portability. However, k8s has several limitations and disadvantages, which teams should consider before adopting this technology. Here, Jonathon Belotti discusses these drawbacks: product engineers must expend time learning how to write applications to k8s; scheduling cronjobs and other batch jobs remains challenging (and expensive!); users must configure and manage clusters (e.g. selecting instance types). He concludes with some ideas on the features that a serverless data platform would provide.