We study how the tasks conducted on the job relate to measures of cognitive skills using data from 18 countries participating in the Programme for the International Assessment of Adult Competences (PIAAC) and from 13 countries that also participate in the International Adult Literacy Study (IALS). We document two main findings. Firstly, individual- fixed effect models suggest that low-educated workers in jobs involving a particular set of basic tasks -say, in numeric rather than reading or ICT tasks- obtain 10% of one standard deviation higher scores in the domain of the PIAAC assessment most related to those tasks than in the rest -say, numeracy relative to literacy or problem-solving scores. The estimates are weaker for workers with a high school or college degree, those with more than 10 years of experience or who are males. Secondly, a synthetic cohort analysis using repeated literacy assessments in IALS and PIAAC indicates that, among the low-educated, long-run increases in the reading task component of jobs correlate positively with increases in cohort-level literacy scores. An interpretation of our findings is that tasks conducted on the job help in building human capital. Under that interpretation, our back-of-the envelope estimates suggest that the contribution of one year of on-the-job learning to skill formation is between a half and a fourth of an extra year of compulsory schooling.