QCD is the only sector of the Standard Model where the exploration of the first levels of complexity, built from fundamental interactions at the quantum level, is experimentally feasible. An outstanding example is the thermalised state of QCD matter formed when heavy atomic nuclei are smashed in particle colliders. Systematic experimental studies, carried out in the last two decades, overwhelmingly support the picture of a deconfined state of matter, which behaves as a nearly perfect fluid, formed in a very short time, less than 5 yoctoseconds. The mechanism that so efficiently brings the initial out-of-equilibrium state into a thermalised system is, however, largely unknown. Most surprisingly, LHC experiments have found that collisions of small systems, i.e. proton-proton or proton-lead, seem to indicate the presence of a tiny drop of this fluid in events with a large number of produced particles. These systems have sizes of 1 fm or less, or time-scales of less than 3 ys. To add to the puzzle, jet quenching, the modifications of jet properties due to interactions with the medium, has not been observed in these small systems, while jet quenching and thermalisation are expected to be controlled by the same dynamics. Present experimental tools have limited sensitivity to the actual process of thermalisation. To solve these long-standing questions we propose, as a completely novel strategy, using jet observables to directly access the first yoctoseconds of the collision. This strategy needs developments well beyond the state-of-the-art in three subjects: i) novel theoretical descriptions of the initial stages of the collision — the first 5 ys; ii) jet quenching theory for yoctosecond precision, with new techniques to couple the jet to the surrounding matter and novel parton shower evolution; and iii) jet quenching tools for the 2020’s, where completely novel jet observables will be devised with a focus on determining the initial stages of the collision.