MVO is the engine behind the efficient-frontier. Feed it expectations and it returns precise weights. The precision is the trap: the output is extremely sensitive to the expected-return inputs, which are the hardest quantity in finance to estimate.
A worked illustration: raise one asset's assumed return from 6.0% to 7.0% and an unconstrained optimiser may move its weight from 15% to 60%, funded by shorting another holding. Nothing about the world changed, only a guess.
Practitioners tame this with constraints (no shorting, maximum weight per asset), shrinkage of the covariance estimate, resampling across many simulated inputs, or the Black-Litterman approach of starting from market weights and tilting only where a view is strong. See naive-diversification for the humble alternative.
Related: efficient-frontier, modern-portfolio-theory, naive-diversification, equal-risk-contribution, monte-carlo-simulation