Everyone's building agents. Is anyone actually good at telling them what to do?
Agents keep failing not because the model is weak but because the instruction is vague — 'handle my outreach' or 'research my competitors' returns generic, confidently wrong work. The people getting good results spell out four things before they start: the goal in one sentence, what 'done' actually looks like, what the agent may never touch, and one example of a good outcome. The skill gap is specifying, not prompting.