When will bots finally do useful work?
Sooner than you think.
First, let's define "useful work."
A bot has done useful work, when it can perform a task currently performed by a human at the same efficiency, reliability, quality and endurance, at or below cost parity with the human wage.
The task does not matter. So long as it is any task currently required in the factory or business.
Clearly the outcome must be the same in quality and reliability, else there is no advantage to using the bot.
Efficiency is important for cost parity as it must not compromise the current production rate or process.
Endurance is extremely important as well. It must be able to perform over an entire shift without failures or glitches, and show up the next day or shift without problem. Proving it can do a task reliably for 10 to 20 minutes does not cut it. An entire shift is required. It can use the same break schedule as humans if mid-shift charging is necessary, but that is all. It needs the durability and reliability to do repeated shifts without interventions. Shifts are marathons, not sprints.
Cost parity is when the operational cost of the bot is at or below the human wage including overhead - FICA, benefits, etc. I will leave it to the accountants to determine the formula for such a calculation.
It's that simple. Once it becomes cheaper to use the bot, and it is dependable, the bot has done useful work.
When that happens, scaling will begin in earnest.