In this work we introduce deep reinforcement learning (DRL) as an environment to develop control strategies for turbulent flows. In the first flow case we consider a turbulent channel, where we compare DRL- based control through blowing and suction with the classic opposition control. Our results show that DRL leads to 43% and 30% drag reduction in a minimal and a larger channel (at a friction Reynolds number of 180), respectively, outperforming opposition control by around 20 and 10 percentage points, respectively. We also show the potential of DRL-based control to reduce the length of a turbulent separation bubble and reduce the drag in a 3D cylinder, in both cases significantly outperforming classical control.