Mob programming is an emergent practice in the industry, attracting increasing attention from the research community. Superficially a simple concept - similar to pair programming, but with more people - many of the same concerns arise as with pair programming, only more so: can it truly be efficient, is it for everybody, and what are the benefits and risks involved? In this paper we show that while research interest in this topic is increasing, much of the published literature to date constitutes experience reports, with a number of implicit differences in how mob programming is practiced. To transform mob programming from an area of avant-garde experimentation into a well understood practice in the toolbox of mainstream software engineering, such implicit differences need to be exposed, analyzed and documented, and the contextual enabling factors must be investigated. In this paper we take an important step in that direction by identifying variations in the practice and providing concrete guidance for creating an organizational and social environment where mob programming can be practiced both effectively and safely. (C) 2021 Elsevier Inc. All rights reserved.