Redis

Most AI leaders will tell you context is critical. Fewer will tell you what's actually broken. This Thursday, Redis CTO Benjamin Renaud and Head of AI Products Simba Khadder sit down to talk about it directly—where context breaks down, why MCP went from overhyped to unavoidable, and what "compounding context" looks like once you've actually built it. Register here: https://lnkd.in/e6e-hpBa 

Picture this: a user searches for "refund policy" but the doc they need is titled "returns and reimbursements." The keyword matching scores it near zero, even though it's exactly what they asked for. Vector embeddings fix that by representing meaning as numbers, so words with similar meanings land near each other in space even when they share no characters. The geometry that results isn't understood in a human sense. Its statistical structure, encoding real relationships (Paris relates to France the same way Tokyo relates to Japan) alongside the messy ones, is baked into the training data. Redis stores those embeddings next to the rest of an app's data, so semantic search runs in the same low-latency system as everything else. Here’s where that geometry needs to live to actually be useful: https://lnkd.in/eB9S5NZH

A stale dashboard is a minor annoyance, but a stale agent… that can be a liability. In 2024, Air Canada's support chatbot told a customer he could apply for a bereavement fare refund after booking, quoting outdated policy from a pipeline that hadn't caught up. A tribunal held the airline liable anyway, ruling it didn't matter whether the information came from a static page or a chatbot. Change data capture is built for exactly this gap. Instead of refreshing context on a schedule, it reacts to changes the moment they commit. Redis Data Integration (RDI) streams those changes into Redis within seconds, so agents work from the current business state instead of the last snapshot. Here's how RDI keeps agent context current, in seconds instead of hours: https://lnkd.in/ejjkUBs5

BREAKING: Redis is joining Merge and PostHog for The Self-Driving Product in NYC! Join us for technical talks about the infrastructure behind AI products that can observe, act, and improve. August 27, 6–8 PM ET RSVP: https://luma.com/gfn0jy68

Mark your calendar: August 27. Redis CTO Benjamin Renaud and Head of AI Products Simba Khadder are sitting down to talk about the gap between what AI leaders believe about context and what they've actually built. 97% say context is critical to their AI goals. Only 4% have built for it. Join us live on August 27 to get the full story.

Retrieval-augmented generation answers questions. It typically doesn't get better at answering them. Ask the same question next month, and the answer won't have improved, because a stateless RAG pipeline doesn't learn from what it retrieved. Memory changes that. It turns a read-only retrieval system into a write-manage-read loop: the agent records what happened, consolidates it, and recalls what's relevant next time. An agent that watches a user correct the same mistake three times can store the fix once instead of relearning it every session. Redis Agent Memory runs that loop with fast enough recall to keep it inside the agent's hot path, not bolted on as a separate service. Here's how to turn your retrieval system into one that actually learns: https://lnkd.in/esTwaB2m

Your context strategy may already be outdated. The way teams are solving context for agents has shifted dramatically. We asked AI and infrastructure leaders what they’re doing now, and one pattern stood out: the teams that can operationalize context are the ones moving agents beyond prototypes and into production. See which approaches are emerging, where organizations are still getting stuck, and how your own strategy compares in the state of context engineering report. Download here: https://lnkd.in/eSsdvpaJ

Redis certifications are here. 👏 When you earn a certification, you send a signal of your infrastructure stack skills—showing you can build production-ready, stable architectures that are fast, scalable, and reliable across development, cloud operations, and software operations. You can choose from: ➕ Redis Associate Developer ➕ Redis Associate Software Operator ➕ Redis Associate Cloud Operator Start your certification journey here: redis.io/certifications Take your exam now and get 100% off with code Earlyaccess100.

Agents don't have an intelligence problem. They have a context problem, and a stale cache is one of the quietest ways that problem shows up. Cache consistency is about shrinking the window where cached data disagrees with the source of truth. For an agent pulling live inventory, account state, or permissions into its context, that window is the difference between a right answer and a confidently wrong one, with no error to flag it. Redis Data Integration handles that layer, streaming source-database changes into Redis so agent context reflects what's true now, not what was true five minutes ago. Read the full guide: https://lnkd.in/eCwxwBej

Getting fresh data into an agent's context is only half the problem. The other half is reading it fast enough that checking for current state on every step doesn't blow the latency budget. Agents re-read context, memory, and tool results at every iteration, and the iteration count adds up fast. Skip the freshness checks to save time, and stale context sneaks right back in. That's the trap teams hit when a read path that's fast enough for a single query becomes the bottleneck once an agent hits it dozens of times per task. Redis Iris keeps context fresh on the write path through change data capture, and is fast enough on the read path that agents can check for current state on every step without paying for it. See how Redis Iris handles your agent's read path or talk to us about keeping context fresh across your AI stack: https://lnkd.in/eHB_pviz