As companies build AI agents into their products, they inevitably need to support integrations that can access and interact with customer data, which is crucial for the AI agents to function effectively. Integrations form the foundation of retrieval-augmented generation pipelines, enabling AI agents to retrieve context needed to take actions on behalf of users, and also provide raw and normalized customer data, supporting time-sensitive processes, such as routing leads or de-provisioning applications. To build integrations with AI agents, companies must navigate the decision of whether to build them in-house or use a third-party provider, implement access control lists, and leverage webhooks to facilitate real-time syncs, while also considering challenges such as hallucinations, security risks, scaling, error handling, and unclear data schemas. There are several options for building integrations, including native builds, embedded integration platform as a service solutions, model context protocol, and unified API platforms, each with its pros and cons, and companies must carefully evaluate these options to determine the best approach for their specific use cases.