AWS adds a context layer for AI agents to read company data
An agent in AWS Context can see only the data that its user can see.
On Wednesday, AWS showed AWS Context at its AWS New York Summit user conference. It puts related company data and business rules in a knowledge graph. This is a map of connected facts that AI agents can examine. AWS says that agents must have this context to give correct results. It is not for general use at this time, and no test results are given.
How it works
AWS Context uses the knowledge graph of Amazon Quick, an AI assistant for employees. It is for all of an organization, not for one person. It learns from use which data agents use together.
An agent can see only the data that its user can see. Governance tools show the data that an agent read, and by which authority.
The proof
No test results are given. Jake Dolezal of McKnight Consulting Group says that it replaces the retrieval pipeline of each team with one context layer. Donald Farmer of TreeHive Strategy says that the identity-aware queries are the best part.
What is not known
Farmer says that the knowledge graph learns from agents. If agents are incorrect, it can copy their errors across the organization, possibly without a person who sees them. Farmer wants to know which feedback signals can correct this. Dolezal says that AWS must also add streaming data for agents.
Insight: Vendors want to control the data layer of agents
On 16 October, AWS released Strands Box, a sandbox that controls which tasks an agent can do. AWS Context controls which data an agent can see. Neo4j and Kong also released context tools in October. This reads as a move by vendors to control the layer between agents and company data. Make sure that you know if you must use only one vendor.
Sources
Posted
