As the job market changes, one way people can maintain or grow their incomes is by pivoting to more in-demands fields. For all the automation and expansion, relatively few people are unemployed (joblessness is a somewhat different boat), and most industries are unable to hire all the people they would like due to lack of appropriate candidates.
This isn't just true of highly paid workers either - there are pressures across wage scales. Some of this is geographic: the right kind of people are available, but not where the company that would hire them is located. This can be made worse by housing policy - the San Francisco Bay Area is a great example of a place where there are needs for people, but while wages are increased they are not increased sufficiently to account for the high price of housing. Some flexibility is limited by industry-specific experience: hiring managers tend to look for people with specific experience in their industry, which will limit the pool even if otherwise credible candidates are available.
Much of the unfilled need is skill-specific though: the requirements are specified in capabilities, not specific industry knowledge. This is much more amenable to training. In some cases the market had come up with offerings - there is sufficient demand and compensation for software development that many for-profit bootcamps (short term training courses) and online courses have developed. They appear to be reasonable effective in both helping people get jobs in that field, and also succeeding in those jobs. This takes some time though - modern software development has been an in-demand career for about 20 years (even incorporating the .com recession) but the environment for these types of new training both required technical developments and attitudinal shifts that occurred as developers moved up to be hiring managers and executives.
As more people have shorter stints in given occupations, such retraining may become a more common event. The Aspen Institute (and others) propose Lifelong Learning Accounts, which would be paid in to by employers and cover retraining during unemployed phases. Along similar lines, the Obama administration pushed for a program that covered dislocated employees, when trade deals results in job loss. It would cover some retraining costs, as well as other social benefits.
I mildly dislike the term "lifelong learning", as it implies without a formal training or education people aren't learning. That is both strictly not true, but also negates a significant amount of firm-specific skill development, and to be honest the kind of industry knowledge that employers seem to value. A lot of this is practical, hard to communicate knowledge about how to get things done, what events within the industry look like and so on. Of course, its value is totally dependent on the continuation of conditions in that specific firm or industry.
In some ways, more traditional education (employer training, universities) give a better leg up in these ares: a good chunk of value of an MBA is the network of fellow students, which somewhat mimics the effect of working in a firm - you are exposed to varying ideas and approaches. Similarly, apprenticeships have a clear line to effectiveness, but by being so focused on a specific job as-it-exists they make new graduates just as vulnerable to dislocations due to trade or automation as long-time workers.
What I don't disagree with is that there will be an increase in the mix of both formal and informal education and training throughout the careers of most people. Examples like the Khan Academy are interesting for describing genuine new approaches - Khan Academy content effectively supplements, rather than replaces, traditional teenage education, and can be used in varying structures which rely more, or less, on the online content. Most of the other MooC providers hew too closely to recreating a university-like environment without moving beyond the lecture+exercise, and significant time investment, model. In part this is driven by the need for equivalent credentials, which is separate failure, and discussion (c.f. Brian Caplan).
Still, given the high volume of content available already, the problem of matching individuals and training together seems pretty significant. For university-level education most countries have developed fairly robust ways of matching people to institutions, and there is still a fair amount of reallocation after students have their first experiences. Adults have to make a similar analysis with much less support, many more choices, higher costs, and more decision factors. Its interesting to see emerging tools here - for example HN Academy collects and ranks computer science courses based on their popularity in the Hacker News community.