If genomes were completely random sequences in a statistical sense, 'overlap-consensus-layout' method would have been enough to assemble large genomes from Sanger reads. In contrast, real genomes often have long repetitive regions, and they are hard to assemble using overlap-consensus-layout approach. De Bruijn graph-based assembly approach was originally proposed to handle the assembly of repetitive regions better.
Comments
This totorial is also very informative http://www.imperial.ac.uk/bioinformatics-data-science-group/resources/software/next-generation-sequencing-ngs-software/de-novo-assembly/