A longitudinal study tracked workers over five weeks as they collaborated with an AI teammate on real tasks. The central finding is consequential: trust and satisfaction changed over time in ways that a one-session test cannot detect. Workers who started skeptical often became more comfortable. Workers who started enthusiastic sometimes grew frustrated. Neither trajectory was visible in the first session alone. The study also found that workers changed their own behavior to fit the AI. They adjusted how they wrote requests, what they delegated, and when they decided to step in. Those accommodations compounded over weeks. A design team that watches only the first session misses the full shape of the relationship. For teams building AI products, the practical consequence is this: the signals that matter for trust and satisfaction are slow signals. A single usability session or a short-run rating gives you a snapshot of a relationship that is still forming. The study argues that evaluating an AI teammate in one session produces a misleading picture, because the measure catches a moment before the working relationship has settled.