The cycling of matter in galaxies between molecular clouds, stars and feedback is a major driver of galaxy evolution. However, it remains a major challenge to derive a theory of how galaxies turn their gas into stars and how stellar feedback affects the subsequent star formation on the cloud scale, as a function of the galactic environment. Star formation in galaxies is expected to be highly dependent on the galactic structure and dynamics, because it results from a competition between mechanisms such as gravitational collapse, shear, spiral arm passages, cloud-cloud collisions, and feedback processes such as supernovae, stellar winds, photoionization and radiation pressure. A statistically representative sample of galaxies is therefore needed to probe the wide range of conditions under which stars form. I will present the first systematic characterisation of the evolutionary timeline of the giant molecular cloud (GMC) lifecycle, star-formation and feedback in the PHANGS sample of star-forming disc galaxies. I will show that GMC are short-lived (10-30 Myr) and are dispersed after about one dynamical timescale by stellar feedback, between 1 and 5 Myr after massive stars emerge. Although the coupling efficiency of early feedback mechanisms such as radiation and stellar winds is limited to a few tens of percent, it is sufficient to disperse the parent molecular cloud prior to supernova explosions. This limits the integrated star formation efficiencies of GMCs to 2 to 10 per cent. These findings reveal that star formation in galaxies is fast and inefficient, and is governed by cloud-scale, environmentally-dependent, dynamical processes. These measurements constitute a fundamental test for numerical sub-grid recipes of star-formation and feedback in simulations of galaxy formation and evolution.