This paper introduces a mechanism to steer long-running in-network computations over ICN, or more specifically, over an NFN network. NFN (Named-Function Networking) is an extension of ICN that performs in-network resolution of expressions instead of mere content retrieval of a single name, as in ICN. One problem NFN faces are timeout events which are not discriminative enough to distinguish between healthy and ongoing computations from blocked threads or from connectivity problems. Moreover, we would like to change parameters of a long-running computations on the fly, to fetch intermediate (approximate) results or to simply cancel/stop/resume a computation. To this end we introduce Request-to-Compute messages (R2C) which generalize keep-alive messages. In this paper we report on experiences with a running prototype (steering a n-body simulation) and relate R2C to PubSub and Rendezvous protocols.