As a layman, I had to think of drugs or other dopamine hits first - actions which increase well-being in the short term but are harmful in the long term.

But what he really talks about seems to be the opposite: Actions which case some mild harm in the short term but increase well-being in the long term. So I guess something like working out or going on a diet or making a downpayment for a house?

Except the failure mode is also counterintuitive: Normally, we tend to overvalue the short-term downsides of those actions and therefore shy away from them, missing out on the long-term benefits. But he talks about a situation where we overvalue the long-term benefits but ignore the short-term and overdo the action until the short-term harm becomes critical.

So, e.g. someone working out, getting muscle-ache - and then working out more to counter the ache - which will only lead to more of it until the workout actually starts to become detrimental to their health.

It's easy to see how this would trip up automated control loops, but I don't really see how this has practical application outside of control theory.