Context is King in AI Agents
summary
Argues that stateful real-time context — not bigger models — determines agent correctness for fraud, AML, cybersecurity and customer support; recommends precomputing live context with pipelines (CDC, Kafka) materialized in ClickHouse via Apache Flink so agents fetch a concise view through an MCP tool instead of pulling raw JSON blobs, lowering LLM token cost and nondeterminism; claims Delta Stream provides connectors, access control and sub-second freshness so agents stay simple and decisions stay correct.
AI-written summary · about
chapters (9)
- 0:00AI agents and context intro
- 3:52Types of context for agents
- 7:26Stateful real-time context
- 14:26Pre-building context benefits
- 17:41Delta Stream fresh context
- 21:52Use cases needing fresh context
- 23:14Anti-money laundering agent
- 26:26Enterprise agents need fresh context
- 27:37Q&A
related talks
15mContext Is the New Code for AI Agents
15mUnderstanding Context Engineering for Agentic Systems
31mBuilding the Context Engine AI Agents Need
31mBuild Intelligent AI Agents With Context Engineering
18mContext Engineering — Reliable AI Instead of Hallucinations
26mWhy More Context Makes Your Agent Dumber and What to Do About It
40mReal-TIme Context Engineering for Agents
27mChallenges of Ambient Agents: Building Governable AI Systems at Scale