Antimicrobial peptides (AMPs) emerge as promising agents against antimicrobial resistance, providing an alternative to conventional antibiotics. Artificial intelligence (AI) revolutionized AMP discovery through both discrimination and generation approaches. The discriminators aid the identification of promising candidates by predicting key peptide properties such as activity and toxicity, while the generators learn the distribution over peptides and enable sampling novel AMP candidates, either de novo, or as analogues of a prototype peptide. Moreover, the controlled generation of AMPs with desired properties is achieved by discriminator-guided filtering, positive-only learning, latent space sampling, as well as conditional and optimized generation. Here we review recent achievements in AI-driven AMP discovery, highlighting the most exciting directions.