Working becomes harder as we grow tired or bored. I model individuals who underestimate changes in marginal disutility - as implied by "projection bias" - when deciding whether or not to continue working. This bias leads to two mistakes. First, they are too pessimistic when they are tired and working is hard, and too optimistic when rested and work is easy. When within-day disutility is convex and individuals face a single task with all-or-nothing rewards (such as passing or failing a test), they initially underestimate the total disutility and start some overly ambitious tasks. As work becomes harder, they perceive the task as less worth completing and may quit. If the deadline for the task is far in the future, such individuals may repeatedly start working, yet quit earlier than anticipated. This can lead to large daily welfare losses. When tasks instead have concave rewards, including piece rates, such individuals work optimally if facing only a single task. But when working on multiple such tasks (for example, studying for two tests with continuous grades), they may mis-prioritize the tasks. In particular, they over-prioritize urgent tasks over important but non-urgent tasks, overestimating how much they will later work on the non-urgent tasks. Second, when tasks can be completed across multiple days, individuals smooth work too little over time. Because they underestimate how much the marginal disutility will increase on better days, they work too much on those days, and overreact to daily differences in opportunity costs, incentives, and productivity.