Redis CEO Rowan Trollope: from in-memory cache to the agent context engine
Key points
Key takeaways from a theCUBE interview, recorded at its New York Stock Exchange studio as part of the NYSE Wired series, with Redis Labs CEO Rowan Trollope (October 2026):
From data platform to agent infrastructure. Trollope framed Redis's next wave: after powering the cloud-mobile era as the in-memory database sitting behind most consumer and enterprise applications — tracking every FedEx package and credit card swipe — customers are now bringing the company agentic problems such as agent memory, orchestration, tracing, reliability and rising token costs.
Iris, the new agent platform. Redis Iris combines an intent router that sends each request to the right model, reserving frontier models for hard problems while cheaper and open-source models handle the long tail; short- and long-term agent memory, including procedural memory of the tools and reasoning a past solution used; agent mesh, built on Redis Streams with an agent-to-agent protocol for coordinating fleets of agents; and agent playbooks for curating skills, FAQs and static knowledge — described as the next generation of RAG pipelines.
Token economics. "It's relatively easy to turn dollars into tokens; it's a lot harder to turn those tokens back into dollars on the bottom line," Trollope said, arguing enterprises are now demanding tight, measurable ROI from agentic projects.
Banks are the proof point. One of the world's largest banks has automated parts of its KYC and AML processes with agents and is already seeing benefits, with the remaining hard problem being evaluation loops — checking that agents made the right decisions, with human review still in the loop. Trollope also said Redis recently closed a deal with the largest bank in the world around agent memory.
The swivel-chair problem. Legacy enterprise data infrastructure is often "tied together by human beings" — visible in contact centres where agents cut and paste between systems; the first hard problem of AI, raw intelligence, is largely solved, while getting the right context into the agent at the right time is the harder challenge enterprises now face.
Curated context over raw data. Agents cannot be handed every database and manual; Redis is working on ontologies within Iris — an area where companies such as Databricks are also active — to describe underlying data to agents and keep context windows curated and cost-effective.
Customers and momentum. Decagon, the customer-service AI company, was cited as racing ahead on agentic workloads, and Trollope noted that new protocols for how agents interact with businesses were being proposed as of the recording, with adoption spreading from the largest banks to the smallest next-generation technology companies.
Read more: theCUBE – NYSE Wired on YouTube
