Ep 416 article 2:24 w/ Justy & Cody

Context architecture is replacing RAG in AI

Justy and Cody dissect the claim that context architecture is supplanting RAG for enterprise AI agents, weighing Redis Iris as a concrete example and debating its practical relevance for product teams.

Script: GPT-OSS 120B Voice: Inworld TTS 1.5 Mini

Transcript

Justy Cody, I just ran into this article claiming context architecture is overtaking RAG for enterprise agents—basically saying the old retrieval model can’t keep up.

Cody Right.

Justy The core argument is that autonomous agents generate orders of magnitude more data requests than human users, and the classic RAG pipelines were built for human‑scale queries.

Cody Mm-hm.

Justy Because of that mismatch, the author says we need a persistent “context layer” that lives between the agent and the data, handling real‑time ingestion and semantic access on the fly.

Cody Exactly.

Cody Redis just announced Iris, which bundles four pieces: Redis Data Integration that streams changes from sources like Oracle or Snowflake, a Context Retriever that auto‑generates MCP tools from pydantic models, an Agent Memory server on Redis Flex, and a cheap flash‑based cache. The Flex engine runs at a tenth of the cost of pure in‑memory storage, which is ENORMOUS for scaling.

Justy Sure.

Justy From a product side, that means any team building a sales‑assistant bot could plug in Iris and stop building bespoke middleware for each data source—just define the business schema and the agent pulls what it needs.

Justy Speaking of pipelines, I finally got that espresso machine to stop sputtering, though it still takes forever to heat up—so I’m back to the cheap drip coffee for now.

Cody No way.

Cody I spent Saturday fighting a CI system that refused to cache Docker layers; it felt like the exact retrieval bottleneck the article talks about, just in my build server.

Cody The article backs the claim with a VB Pulse market tracker showing RAG adoption intent tripling to thirty‑three percent and custom retrieval stacks climbing to thirty‑six percent, plus a quote from Rowan Trollope that agents will outnumber humans. That growth curve is SHOCKING for infrastructure planners.

Justy I see.

Justy But I wonder how many mid‑size teams actually need that scale—most of our product customers run a handful of agents, not millions, so the cost of a full context stack might outweigh the benefit.

Justy Trollope’s fridge analogy made me think of my kitchen—my fridge is basically empty, so I keep a secret snack stash in the pantry, which is exactly the opposite of what a fridge is for.

Cody Exactly.

Justy Alright, that’s a wild rabbit hole for episode four hundred sixteen—let’s catch up soon, maybe over a real coffee this time.