AI Universe · The AI application (agent harness)
The AI application (agent harness) zoom 2
What surrounds the model call? 7 concept buckets. Each one is a job you can learn once; underneath it, who does that job on each platform.
Concept
Control loop
Ask the model, run the tools it asks for, go round again, stop safely.
Physical: who does this job, where
| Open source | plain Python, LangGraph, Pydantic AI, OpenAI Agents SDK, CrewAI |
|---|---|
| Azure | Microsoft Agent Framework, Foundry Agent Service |
| AWS | Bedrock Agents, Strands, AgentCore runtime |
| GCP | Agent Development Kit, Vertex AI Agent Engine |
| Vendors | Claude Agent SDK |
Concept
Context assembly
Building exactly what the model reads on every call, within a budget.
Physical: who does this job, where
| Open source | your own code, LlamaIndex |
|---|---|
| Azure | Foundry prompt flow |
| AWS | Bedrock prompt management |
| GCP | Vertex prompt management |
| Vendors | provider context tools |
Concept
Tools and protocols
Letting the model act in the world, safely and typed.
Physical: who does this job, where
| Open source | MCP SDKs, FastMCP, A2A SDK |
|---|---|
| Azure | APIM MCP, Foundry tools |
| AWS | AgentCore Gateway |
| GCP | Vertex AI extensions, ADK tools |
| Vendors | MCP servers from SaaS vendors |
Concept
State and memory
Remembering the conversation and facts, across turns and sessions.
Physical: who does this job, where
| Open source | PostgreSQL, Redis, Mem0, Letta |
|---|---|
| Azure | Cosmos DB, Azure PostgreSQL |
| AWS | AgentCore Memory, DynamoDB |
| GCP | Agent Engine memory, Firestore |
| Vendors | Zep |
Concept
Guardrails and policy
Checks before and after every step: what may go in, come out, and run.
Physical: who does this job, where
| Open source | your own rules, Guardrails AI, NeMo Guardrails, Llama Guard |
|---|---|
| Azure | AI Content Safety, Prompt Shields |
| AWS | Bedrock Guardrails |
| GCP | Model Armor |
| Vendors | Lakera |
Concept
Durable work and people
Waiting hours for a person, surviving restarts, never paying twice.
Physical: who does this job, where
| Open source | DBOS, Temporal, Restate |
|---|---|
| Azure | Durable Task Scheduler, Logic Apps |
| AWS | Step Functions |
| GCP | Workflows |
| Vendors | Inngest |
Concept
Identity
Who is asking, and whose authority the agent acts with.
Physical: who does this job, where
| Open source | Keycloak, PyJWT |
|---|---|
| Azure | Entra ID, managed identity |
| AWS | Cognito, IAM, AgentCore Identity |
| GCP | Identity Platform, IAM |
| Vendors | Okta, Auth0 |