A study of 73 000 posts from OpenClaw, a community for people who run AI agents, finds that humans do not hand work to an agent in a single move. Each person shapes the task around personal values, risk tolerance, and the specific stakes involved. The researchers call this value-sensitive design: humans embed what matters to them before the agent starts, not after. The configuration screen before a run is where a human decides what the agent is allowed to want on their behalf. The study applies Value Sensitive Design, a framework for surfacing human values inside technical systems, to real agent use at scale. Three pressures emerge. Humans want an agent to respect their own ethical lines, not a generic default. They want to narrow the agent's reach before it starts. They also want to see where the agent's knowledge came from so they can judge how much to trust the plan. For product teams, this evidence points at a gap in most interfaces. The configuration screen before a run is treated as logistics: pick a model, set a budget, press go. OpenClaw's data suggests humans arrive at that screen carrying values and wanting to express them. A team designing that screen should let the human state what matters, not only set parameters.