Will robots eliminate human jobs? Experts at the McKinsey Global Institute have long argued that’s the wrong question to ask about automation and the future of work. The reason: it fails to recognize the fundamental distinction between “jobs” and “tasks.”

Most jobs involve performing a variety of different tasks. An occupation like “travel agent” might involve a host of skills that are easy for machines to match: knowledge of geography or an ability to understand airline and train schedules. But it also requires other, hard-to-automate talents such as intuiting customers’ hopes and dreams and selling an appropriate travel package.

McKinsey analysts argue that, over the next decade, robots will take over many tasks—perhaps even half of all the things humans now get paid to do. But they see few occupational categories in which robots are likely to take over entire jobs. McKinsey’s research suggests that in years to come, humans will collaborate more and more closely with machines but not get pushed out of the workplace entirely.

Using data from the U.S. Bureau of Labor Statistics and O*Net, MGI recently conducted a detailed analysis of more than 2,000 work activities for more than 800 occupations. Their goal: to assess the technical feasibility, using currently demonstrated technologies, of automating three groups of occupational activities: those that are highly susceptible, less susceptible, and least susceptible to automation. In a recent article in the McKinsey Quarterly, MGI’s Michael Chui, James Manyika, and Mehdi Miremadi described some of the conclusions of that analysis. The whole article is worth reading.

You can get a sense of MGI’s analysis of which occupational categories are most and least vulnerable to automation from the graphic below. It’s a matrix depicting eight types of occupations across 19 different economic sectors. For each job box, the wider the color bars, the larger the percentage of time on the job spent on activities that can be automated. Yellow, green, and blue color bars indicate tasks that are highly automatable, while orange and red bars indicate tasks that are hard to automate. The implication: look for jobs with the skinny red lines and steer clear of the ones with the fat blue bars.