AWS view · documentation checked 1 November 2026
Assemble capabilities around the business decision.
AWS documents building blocks for models, agent runtime, retrieval and workflow control. The CAE interpretation connects them to enterprise context, human authority and systems of record.
An original candidate architecture
- Experience and integrationExisting business channels → Amazon API Gateway and authorised application adapters.
- Context and meaningExisting records → Glue Data Catalog / Lake Formation and Bedrock Knowledge Bases where suitable. Business owners maintain meaning and source quality.
- Reasoning and workflowAmazon Bedrock models and AgentCore runtime; Step Functions for explicit workflow coordination where suitable.
- Authority and effectConfigured identities, AgentCore Gateway Policy on mediated traffic, human decision application and backend transaction checks.
- Evidence and operationSelected AgentCore operational capabilities plus customer decision evidence, outcome measures, alerts and recovery.
This diagram is a proposed assembly. All arrows denote integration work to design and test, not an existing connection.
What documentation supports
Amazon Bedrock supplies managed model access. AgentCore documents runtime, memory, gateway and identity components that can be selected separately. Knowledge Bases supports retrieval from configured information sources. These capabilities do not establish the accuracy or permitted use of customer information.
Gateway Policy evaluates traffic passing through its gateways. Protect other tool and API paths separately. Step Functions coordinates state-machine workflows; its human approval example demonstrates an application pattern rather than supplying the organisation’s approval policy.
API Gateway provides API-management capabilities. Lake Formation governs supported data access alongside catalogue metadata. Neither defines business concepts or grants a business mandate.
Controls and deployment choices still to make
- Choose the region and verify each feature’s availability and release stage; the overview does not establish a regional feature matrix.
- Select model eligibility, approved processing locations, retention, quotas and budget limits.
- Implement reviewer identity, exact-action approval, expiry, idempotency, transaction checks and recovery.
- Configure source permissions and check retrieval against representative permitted and denied cases.
- Attribute model, runtime, workflow, storage, egress, support and human-review costs to the operation.
- Verify backup and recovery against the service objectives using AWS architectural guidance.
No AWS account, customer data, paid service, runtime or integration is provisioned by this website. No benchmark, partnership, certification or compliance outcome is claimed. See the full capability comparison for the customer-owned responsibilities across all three clouds.
YOUR ENTERPRISE. YOUR NEXT CHAPTER.
Turn complexity into
a clear next move.
Bring a business question. We can discuss a discovery scope and the decisions it should support.
Start a discovery conversation