Chemical mechanisms form the core of atmospheric models to describe degradation pathways of pollutants and ultimately inform air quality and climate policy makers and other stakeholders. The accuracy of chemical mechanisms relies on the quality of their input data, which originate from experimental (laboratory, field, chamber) and theoretical (quantum chemistry, theoretical kinetics, machine learning) studies. The development of robust mechanisms requires rigorous and transparent procedures for data collection, mechanism construction and evaluation, and creation of reduced or operationally defined mechanisms. Developments in analytical techniques have led to a large number of identified chemical species in the atmospheric multiphase system which have proved invaluable for our understanding of atmospheric chemistry. At the same time, advances in software and machine learning tools have enabled automated mechanism generation. We discuss strategies for mechanism development, applying empirical or mechanistic approaches. We show the general workflows, how either approach can lead to robust mechanisms and that the two approaches complement each other to result in reliable predictions. Current challenges are discussed related to global change, including shifts in emission scenarios that result in new chemical regimes (e.g. low NO scenarios, wildfires, mega/gigacities) and require the development of new or expanded gas- and aqueous-phase mechanisms. In addition, new mechanisms should be developed to also target oxidation capacity, and aerosol chemistry impacting climate, human and ecosystem health.