The increasing burden of multidrug-resistant bacteria affects the management of several infections. In order to prescribe adequate antibiotics, clinicians facing severe infections such as hospital-acquired pneumonia (HAP) need to promptly identify the pathogens and know their antibiotic susceptibility profiles (AST), which with conventional microbiology currently requires 24 and 48 h, respectively. Clinical metagenomics, based on whole genome sequencing of clinical samples, could improve the diagnosis of HAP, however, many obstacles remain to be overcome, namely the turn-around time, the quantification of pathogens, the choice of antibiotic resistance determinants (ARDs), the inference of the AST from metagenomic data and the linkage between ARDs and their host. Here, we propose to tackle those issues in a bottom-up, clinically driven approach.