The article emphasizes the need for an "agent contract" to address the disconnect between teams and systems in AI development. It identifies the same issues that data contracts aimed to solve: misalignment between data producers and consumers, which often leads to reliability problems. AI agents depend on data and tools managed by different teams, and when those inputs change unpredictably, the output suffers. Establishing clear agreements between teams can help mitigate these risks, enhancing trust and reliability in AI systems. To create an effective agent contract, the article outlines core principles such as working backwards from consumer requirements and agreeing on standards and definitions. It stresses that the quality of outputs is directly tied to the quality of inputs, and that leadership should foster collaboration between AI and data engineers. The contract should include provisions related to knowledge management, data engineering, IT, and AI engineering, detailing everything from data classification and schema to tool availability and performance metrics. Each change in these areas would require notification and potential amendments to the contract. Establishing this framework is essential for managing the complexities of AI systems as they integrate into business workflows. The article argues that reliability isn’t just a technical issue; it’s fundamentally about organizational coordination and communication. If teams fail to align, they risk facing challenges like silent failures and unclear ownership, which can disrupt operations. The effort to develop reliable AI agents is significant, but it’s necessary for success in increasingly complex environments.