As software companies scale agentic AI from isolated pilots to production-ready use cases, the hardest part is rarely the model itself, but the data around it. That’s exactly the bottleneck that Redis Iris, a new context and memory layer for agentic AI, hopes to address. Launched in May, the product is designed to close the gap between an AI agent and the business systems it needs to draw from to function optimally. This way, the agent can retrieve live information, remember previous interactions, and work across otherwise fragmented enterprise data sources.
The launch of Redis Iris is just the latest example of a major trend in data management. Until recently, most attention has focused on retrieval-augmented generation (RAG). However, despite being the dominant pattern in enterprise generative AI for several years, it’s now starting to show its limitations when it comes to getting the right business context at the right moment, without creating new risks around speed, accuracy, or governance—all of which are even more important in the agentic era.
Redis is positioning Iris as an answer to that problem, stating that the platform includes tools for context retrieval, agent memory, data integration, caching, and search. Two of those tools, Redis Context Retriever and Redis Agent Memory, are completely new. Redis Data Integration, meanwhile, is now generally available and is designed to help teams keep data flowing from existing systems like databases and warehouses.
Redis shared an example of a customer service agent tasked with answering a question about a delayed order. In such cases, the agent might need information from an order system, a shipping provider, a customer profile, a support ticket, and a policy document. The problem is, that information tends to live in different systems. An employee might be able to navigate that mess, albeit slowly. However, with the right data infrastructure, an AI agent can access the same context in mere seconds, while predefined guardrails ensure it accesses only what it needs to get the job done.
The data layer has always been crucial in AI, but it’s even more important in agentic systems, given they’re intended to function autonomously. A simple search through stored documents might be enough for a basic chatbot, but not for an agent expected to carry out complex multi-step tasks and constantly refresh its understanding as circumstances change. For software companies, that changes the data-management conversation, because it’s no longer just about where data is stored, but how fresh it is, how clearly it’s tagged, and how safely it can be exposed to autonomous systems.
As for the launch of Redis Iris and other products like it, there’s a clear commercial angle too. Most companies aren’t going to replace their entire systems of record just to support AI agents, at least not any time soon. What they do need is a way to make those data estates usable by agentic AI without forcing a full migration.
For software managers, that means ownership will need to be clearer, with product, data, platform, and security teams all having a stake in whether and how agents can use business information reliably. Redis Iris isn’t the final answer to that challenge, but it is a sign of a new reality where enterprise AI work is moving away from model demos, and toward the data foundations that make agents useful in the real world.