Business processes are facing increasing pressure to quickly and flexibly adapt to changes in the process context. Moreover, microservices are becoming increasingly popular as an architectural style for partitioning business logic into small services accessible with lightweight mechanisms, leading to increasing pressure for a more dynamic integration of information services with processes. Process-aware information systems must thus increasingly incorporate the ability to react to unforeseen changes during process enactment, facing difficulties in pre-modelling all the possible process variations and enactment circumstances for larger process models. This paper presents Microflows, an automatic lightweight declarative approach for the workflow-centric orchestration of semantically-annotated microservices using agent-based clients, graph-based methods, and the lightweight semantic vocabularies JSON-LD and Hydra. The evaluation results show the approach's potential in lightweight resource utilization, investigates its scalability, and compares its automation to common manual workflow modeling and enactment.