We identify 42 candidate groups lying between 1.8<z<3.0 from a sample of 3502 galaxies with spectroscopic redshifts in the zCOSMOS-deep redshift survey within the same redshift interval. These systems contain three to five spectroscopic galaxies that lie within 500kpc in projected distance (in physical space) and within 700km/s in velocity. Based on extensive analysis of mock catalogues that have been generated from the Millennium simulation, we examine the likely nature of these systems at the time of observation, and what they will evolve into down to the present epoch. Although few of the member galaxies are likely to reside in the same halo at the epoch we observe them, 50% of the systems will eventually bring them all into the same halo, and almost all (93%) will have at least part of the member galaxies in the same halo by the present epoch. Most of the candidate groups can therefore be described as proto-groups. An estimate of the overdensities is also consistent with the idea that these systems are being seen at the start of the assembly process. We also examine present-day haloes and ask whether their progenitors would have been seen amongst our candidate groups. For present-day haloes between 10^14-10^15Msun/h, 35% should have appeared amongst our candidate groups, and this would have risen to 70% if our survey had been fully-sampled, so we can conclude that our sample can be taken as representative of a large fraction of such systems. There is a clear excess of massive galaxies above 10^10Msun around the locations of the candidate groups in a large independent COSMOS photo-z sample, but we see no evidence in this latter data for any colour differentiation with respect to the field. This is however consistent with the idea that such differentiation arises in satellite galaxies, as indicated at z<1, if the candidate groups are indeed only starting to be assembled.