On June 19 at AWS Summit NYC, Amazon dropped something that changes the agent infrastructure game.
AWS Context — a managed knowledge graph that gives every agent in your stack a shared, queryable understanding of your business data. Not a document store. Not RAG on steroids. A knowledge graph.
They also shipped:
- Amazon Bedrock AgentCore — a managed runtime for agents that handles state, memory, and orchestration.
- AWS DevOps Agent — GA version of their autonomous ops agent.
- Amazon Connect Decisions, Talent, Customer — four agentic AI products for supply chains, hiring, and CX.
This is the big three (AWS, Microsoft, Google) now all-in on agent infrastructure within a 90-day window. And it creates a new problem that every business deploying agents needs to solve.
Why Context Changes the Verification Game
Before Context, your agents had disjointed data sources. Agent A knew your product catalogue. Agent B knew your pricing. Neither knew the other existed, and neither could tell you when their knowledge diverged from reality.
Context fixes the knowledge problem. But it creates a verification problem:
If every agent shares a knowledge graph, who verifies that the knowledge graph is correct?
Here's what AWS Context doesn't do:
1. It doesn't verify agent actions against real-world outcomes. The knowledge graph knows your return policy. It doesn't know whether an agent actually followed it.
2. It doesn't create an immutable proof trail. A knowledge graph is mutable by design. Who changed what, when, and was it authorised?
3. It doesn't check agent-to-agent claims. When Agent A tells Agent B "the customer is premium tier," nothing verifies that assertion independently.
These are the gaps that the Attest proof layer fills — signed receipts for every agent action, independent verification of claims, and an audit trail that lives outside any single knowledge graph.
What This Means for Agents in Your Business
If you're running agents today, AWS just gave you better infrastructure. But better infrastructure means more agent actions, faster. And more actions means more claims that need independent verification.
The businesses that thrive in the agent era won't just be the ones that deploy agents. They'll be the ones that can prove their agents are working correctly, to customers, regulators, and their own operations teams.
| Before Context | After Context | |---------------|--------------| | Agents with fragmented knowledge | Single shared knowledge graph | | Manual data plumbing | Managed infrastructure | | No audit trail for agent decisions | Graph-level queries | | Verification = impossible | Verification = the differentiator |
The knowledge graph is the operating system. The proof layer is the security camera. You need both.
---
This is the third major agent infrastructure announcement this quarter, joining Microsoft Agent 365 and Google Cloud's Vertex AI Agent Builder. The infrastructure race is over. The verification race just started.