The simulation of Monte Carlo (MC) events is a crucial task and an indispensable ingredient for every physics analysis. Geant4 is the state-of-the-art tool used for detailed simulations of the ATLAS detector, which however requires large CPU resources. To reduce the CPU needs, which in turn enables the production of higher statistics MC samples, ATLAS has developed a strong program to replace parts of the simulation chain by fast simulation tools. These developments pave the way towards High Luminosity LHC when resources will be even scarcer. Among those tools is AtlFast3, which utilises a combination of Generative Adversarial Networks (GANs) and sophisticated parametrisations for the fast simulation of showers in the electromagnetic and hadronic calorimeters. For the Run 3 MC campaign, various improvements of AtlFast3 were developed, for example a refinement and extended usage of the GANs and a better model of the punch through of showers into the muon system. Consequently, the performance of AtlFast3 in Run 3 is better than ever. ATLAS also aspires to use fast simulation in the inner detector. FATRAS is a tool that approximates particle interactions with the material through physics formalisms. An integration of FATRAS with the experiment-independent common tracking software (ACTS) is also in development. Track overlay is a technique to speed-up the production of MC samples that include additional interactions (pile-up) aside the hard-scatter interaction. The idea is to reconstruct pile-up tracks before they are merged with the hard-scatter, which reduces CPU needs. Machine learning techniques are used to ensure this method can even be applied in dense tracking environments. This talk will discuss the status of the development of these tools as well as their performance in terms of physics modelling and computing resources.