This week Redis launched Iris, a context and memory platform designed specifically for enterprise AI agent workloads. It is a product that addresses something most AI deployment teams have quietly been hitting for months: the retrieval architecture that worked fine for humans does not work at the speed and volume that agents demand.

The announcement lands alongside fresh market data showing the shift is already underway. According to VentureBeat’s Q1 2026 VB Pulse RAG Infrastructure Market Tracker, buyer intent to adopt hybrid retrieval systems tripled between January and March 2026, jumping from 10.3% to 33.3%. Retrieval optimisation displaced evaluation as the top enterprise investment priority for AI infrastructure for the first time.

What Redis Iris Actually Is

Iris sits between an AI agent and the business data it needs to act. That sounds simple, but it is solving a structural problem that has been underappreciated.

Traditional RAG (Retrieval Augmented Generation) was built around how humans use search. You write a query, it finds relevant documents, you get a response. That pipeline works when the system handles a few hundred requests per hour and the data changes slowly.

Agents work differently. A single agentic workflow can make thousands of data requests as it plans, checks, decides, and executes. The data it needs can change mid-task. The context from earlier in the conversation has to travel with it. Classic RAG pipelines were never designed for that volume or that state management requirement.

Iris has three core components:

Redis Context Retriever (in preview): A semantic interface that auto-generates MCP tools from existing business data models, letting agents query live business data without custom integration work for each data source.

Redis Agent Memory (in preview): A persistent memory layer that lets agents carry context across sessions and multi-step workflows, solving the problem of agents that “forget” previous work when a task runs long or resumes later.

Redis Data Integration (generally available as of May 18): A pipeline that keeps data flowing from external systems into Redis in near real time, so agents are working with current information rather than stale snapshots.

The underlying storage engine, Redis Flex, runs 99% of data on flash storage at roughly one-tenth the cost of purely in-memory systems. That matters for enterprises running high-volume agent workloads where in-memory costs at scale would otherwise make the economics unrealistic.

Redis CEO Kevin Scott Trollope describes the architecture shift as an inversion: “It is just a flip to let the agent pull the data instead of presupposing and stuffing it into the pipeline.” That framing captures why this is a bigger deal than it sounds. Most RAG systems require developers to predict what context an agent will need and push it in ahead of time. Iris lets the agent pull what it needs, when it needs it, from current data.

Why This Matters Right Now

Forty-three percent of enterprise AI agent stacks already use Redis somewhere in their runtime layer. Redis is not a startup making a market claim here. It is an infrastructure company that is already embedded in the systems organisations are trying to scale, expanding into the context layer those systems need as agents take on more complex work.

The broader picture is that enterprise AI deployment has moved past the model selection problem. Most organisations have access to capable models. The bottleneck is now infrastructure: how do you give an agent reliable, governed, low-latency access to the business data it needs to do useful work without rebuilding your entire data stack?

Context architecture is the answer the market is converging on. The Q1 2026 data is a leading indicator that enterprises are not waiting for this to be solved theoretically. They are already changing procurement priorities to address it.

What This Means for Business

If you are building AI agents internally, the retrieval layer deserves more attention than it usually gets in early-stage deployments. Pilots that run on small, static datasets often look great. The same architecture at production scale, with live data and multi-step workflows, frequently fails. Redis Iris is one answer to that gap, but the broader point is to stress-test the retrieval layer before you commit to it.

If you are evaluating enterprise AI platforms, ask vendors specifically about how their agents handle state management and real-time data access. Demos typically use curated data. Production runs on the messy, constantly changing data your business actually generates. Those are very different problems.

If you are a data team supporting AI initiatives, context architecture is going to become a core part of your infrastructure roadmap. The model runs. The data pipelines and memory systems are what make or break production-grade agents.

The memory problem has been one of the less-discussed reasons enterprise AI agents underperform in production. Redis Iris is a direct response to it, and the market data suggests the timing is right.

---

If you’re deciding where to start with agents, start here. The free Working With Claude field guide walks through the ecosystem, Claude Code, and a real rollout plan. Get your copy.