Context Engineering Emerges as Top Priority for AI Agent Success

Context is the bottleneck, and unfortunately, it's not something simple to solve.

Here is the way I think about "the right context" (the 4 ingredients I'm always looking for):

1. Context has to be navigable: You want agents to understand and traverse relationships in the context. This is much more useful than thinking of context as "chunks of information".

2. Context has to be fast: You don't want to spend your latency budget retrieving and building your context.

3. Context has to be fresh: You need to decide when your context is stale and when to refresh it.

4. Context has to compound: The more you use the system, the better it should get.

Redis released Iris, a brand new agent context and memory platform that you can use to power smart agents:

First, they released the Redis Context Retriever, a semantic layer that sits over your live data.

Instead of getting top-k chunks from a vector store, the Context Retriever lets the agent traverse entities and find related information. For instance, Customer → Invoice → Product.

This is closer to graph traversal than regular retrieval.

Second, they released the Redis Agent Memory, a managed short and long-term memory for agents.

This will handle embeddings, retrieval, summarization, and durable state across sessions.

Every team that's building agents is now reinventing this, and most do it badly.

Third, they made Redis Data Integration generally available. This allows you to continuously sync from Postgres, MySQL, Oracle, warehouses, and document stores into Redis.

This is how you can keep your context fresh.

When you combine these three with Redis Search, you can handle semantic, hybrid, keyword, and structured queries with very low latency.

Thanks to the team for partnering with me on this post.