My proposed talk is based on the experience implementing Swarm.js and swarmdb prototypes, as well as new experimentation. My focus is on finding techniques to organise a CRDT system’s inner workings. The super-powers of CRDT are known: offline writes, real-time sync, collaborative abilities, extreme throughput. The top expense is the necessity to break some established practices of software development. Hence, the goal of the work is to realise the superpowers while minimising the breakage.