We develop a dynamic model of attorney-client agency combining hidden information and hidden action. The attorney observes case quality but cannot credibly communicate it; the client learns through Bayesian updating from the absence of revealed settlement opportunities. Strategic concealment by the attorney slows this learning process. The attorney’s decision to conceal opportunities is characterized as a real option, trading off immediate gains against the value of waiting for lower-effort opportunities. Contingency fees mitigate this option value but require sharing potential gains with the attorney. Nash bargaining predicts moderate contingency rates. Combining ex-ante screening with dynamic learning generates predictions consistent with empirical patterns.