Unemployment is a waste of the economy's most important resource, and the presumption against it is strong. But "always" invites examination, and there are cases where it does not hold.
The case that it is always bad rests on waste and its permanence. An unemployed worker produces nothing, so the economy operates below its capacity and the output foregone is never recovered. The individual loses income, skills and health, and the damage to their earnings persists long after they return to work. The government loses tax revenue and pays more in benefits, so it has less to spend on the investment that would prevent the problem recurring. And if the unemployment lasts, skills decay and investment stops, so the economy's capacity itself falls and the cost becomes permanent rather than temporary.
The case that some unemployment is not bad begins with the observation that not all unemployment is the same. Frictional unemployment is the time workers spend searching, and some of it is productive: a worker who takes three weeks to find a job that suits their skills produces more, for longer, than one who takes the first offer on the first day. Matching badly costs an economy more than searching briefly. A labour market with essentially no frictional unemployment would be one in which people cannot move between jobs, which is a symptom of rigidity rather than health.
Unemployment also has effects that some regard as useful. A pool of available labour lets an expanding industry recruit without bidding wages up, so the economy can grow without generating inflation — and an economy with no available workers cannot reallocate labour to new industries at all, since every worker is already committed. That reallocation is how an economy adapts, so a small margin of available labour is part of how structural change happens without inflation.
My judgement is that unemployment is almost always bad, and that the exception is confined to the frictional minimum — but that this exception matters because it determines what a government should aim at. If all unemployment were bad, the correct target would be zero, and a government would keep stimulating demand until it got there. Because a few per cent is the unavoidable and partly productive by-product of a labour market in which people move between jobs, pursuing zero means stimulating an economy that has already absorbed its available labour, which produces inflation and no additional employment.
Beyond that minimum the argument collapses entirely. Cyclical unemployment is pure waste — workers willing, employers unwilling, no vacancies. Structural unemployment is pure waste with a slower cure. Seasonal unemployment is a real cost that can be reduced by diversification. None of these has any compensating benefit, and the "available labour restrains wages" argument is a benefit to employers rather than to the economy: the same restraint could be achieved by raising productivity, without anyone being out of work.
So the honest answer is that unemployment is bad except for the small frictional component, and that the practical importance of the exception is that it defines full employment as something other than zero.