Problem definition and refinement.
Assuming that humans remain as decision-makers when it comes to defining what high-level problems should be solved, e.g. defining what type of life we want to enjoy, what we would like the world to look like, what makes us happy, etc, humans will need to specify (and most importantly give feedback on) those specific goals, objectives, and to keep refining and adapting them to evolving preferences and the evolving human condition.
One could easily argue that the above constitutes work (as it does today), and can definitely have a meaningful and significant impact in the economy and the general computation & work done by our society. Note that we call the above programming or machine learning (~programming from data) in today's world.
Note that nowhere is written that our goals in life will remain fixed. In fact, the opposite is quite more likely. But perhaps most importantly, inferring (estimating) what constitutes joy, and our goals and preferences with varying time horizons, and as we evolve over time, may be an unsolvable problem, and thus require infinite time and our continuous feedback and not-necessarily-conscious guidance.