We study a repeated principal-agent model with transferable utility, where
the principal's evaluation of the agent's performance is subjective. Our
focus is on equilibria which are robust to the addition of small privately
observed shocks to the payoffs. Existing constructions of positive-effort
equilibria are not robust to such payoff shocks. Allowing for simultaneous
cheap-talk announcements makes some effort sustainable in a robust
equilibrium, and payoffs can be arbitrarily close to fully efficient ones if
players are sufficiently patient. In contrast to the existing literature,
our near-efficient equilibria exhibit realistic features: the bonus size is
reasonable, the threshold for being paid a bonus is non-trivial, and the
base wage need not be negative.