This paper exploits novel data on the degree of automatability of approximately 100 skills and abilities collected through an original survey of experts in AI, and link them to occupations using information on skill and ability requirements extracted from O*NET. Similar to previous studies, this allows gauging the number of jobs potentially affected by automation and the workers who are most at risk of automation. The focus on the automatability of skills and abilities as opposed to entire occupations permits a direct assessment of the share of highly automatable and bottleneck tasks in each occupation. The study finds that thanks to advances in AI and robotics, several high-level cognitive skills can now be automated. However, high-skilled occupations continue to be less at risk of automation because they also require skills and abilities that remain important bottlenecks to automation. Furthermore, jobs at highest risk of automation will not disappear completely, as only 18 to 27% of skills and abilities required in these occupations are highly automatable. Rather, the organisation of work will change and workers in these jobs will need to retrain, as technologies replace workers for several tasks.