> Put briefly: give a rational agent more than one thing to do, and they will only do the most profitable thing for them to do. To avoid this problem you need perfectly equal compensation of their alternatives, but that's flawed too, because you rarely want an agent to divide their time exactly into equal shares.
I would argue the system is working as intended. Contrary to your assertions, you don't want employees spreading effort like peanut butter, you want to focus them on executing one or two things quickly and getting value out of that quickly. Instead of launching 12 features a year from now, I'd rather launch 1 feature a month.
> you cannot improve a process that you have not first measured and then stablised.
There is of course, a certain amount of reasoning under uncertainty involved. One of the lessons many folks learn from a/b testing and OKRs is just how hard it is to actually make a difference, and folks need practice calibrating.
> Contrary to your assertions, you don't want employees spreading effort like peanut butter, you want to focus them on executing one or two things quickly and getting value out of that quickly.
That's not quite what I was driving at. Optimisation is made on the measurement. Measurement is only necessary because the Agent is not perfectly observable, there is an information asymmetry between Principal and Agent.
That's why Austin's model is so helpful. There are many things that must be done in order to best satisfy the Customer. Some of those are measurable, some are less measurable. But a rational Agent looks at any basket of measurements and will optimise for one of them: the one that pays best.
It's not enough to say "just this one feature and no peanut butter please". You have to define what the one feature is. You have to provide an exact measure for it. Agents can then either optimise honestly, or they can go further and optimise fraudulently. If honestly, the Principal realises that they actually need a basket of values to be optimised. But then they need to apply equal compensation, because the Agent will simply ignore any measurement that doesn't maximise their results.
I believe measurement is useful. But I also believe that connecting it to even the whiff of reward or punishment is beyond merely futile and well into being destructive.
I would argue the system is working as intended. Contrary to your assertions, you don't want employees spreading effort like peanut butter, you want to focus them on executing one or two things quickly and getting value out of that quickly. Instead of launching 12 features a year from now, I'd rather launch 1 feature a month.
> you cannot improve a process that you have not first measured and then stablised.
There is of course, a certain amount of reasoning under uncertainty involved. One of the lessons many folks learn from a/b testing and OKRs is just how hard it is to actually make a difference, and folks need practice calibrating.