Browse AWS Certification Guides

Implementation and Integration

Study AIP-C01 agents, tool integration, model deployment, FM APIs, enterprise systems, workflow orchestration, routing, and GenAI application patterns.

This domain is where AIP-C01 stops being a GenAI vocabulary test and becomes a systems-design exam. You need to connect foundation models to tools, deployment targets, APIs, enterprise systems, and development workflows without losing safety, resilience, observability, or maintainability.

What AWS is explicitly testing

As of May 11, 2026, the current AWS Documentation exam guide splits this domain into five tasks:

  • Task 2.1: implement agentic AI solutions and tool integrations
  • Task 2.2: implement model deployment strategies
  • Task 2.3: design and implement enterprise integration architectures
  • Task 2.4: implement FM API integrations
  • Task 2.5: implement application integration patterns and development tools

That split matters on the exam. If the problem is about memory, state, tool choice, MCP, approvals, model coordination, or agent workflow, it belongs in task 2.1. If it is about where and how the FM or app runtime should be deployed, it belongs in task 2.2. If it is about CRM, contact center, document systems, business apps, identity federation, or knowledge-system integration, it belongs in task 2.3. If it is about Bedrock API behavior, streaming, asynchronous processing, fallback, retries, rate limiting, or FM invocation contracts, it belongs in task 2.4. If it is about Step Functions, EventBridge, Prompt Flows, CI/CD, Amazon Q Developer, API interfaces, or application-development patterns, it belongs in task 2.5.

Current weight in the exam guide

AWS currently weights this domain at 26% of scored content.

Official task compression

2.1 Agentic AI and tool integration is about deciding when autonomous tool choice is justified, how memory and state are managed, how tool boundaries are enforced, and where human approval or deterministic orchestration should replace full autonomy. Start with 2.1 Agentic AI Solutions and Tool Integrations.

2.2 Model deployment is about choosing the right serving path for application needs: on-demand invocation, provisioned throughput, SageMaker AI endpoints, containerized model hosting, hybrid patterns, and model cascading. Start with 2.2 Model Deployment Strategies.

2.3 Enterprise integration is about safely connecting GenAI features to existing applications, data sources, identity systems, events, legacy systems, and compliance-constrained environments. Start with 2.3 Enterprise Integration Architectures.

2.4 FM API integration is about invocation contracts: synchronous versus asynchronous calls, streaming, request validation, retry and fallback behavior, rate limiting, model routing, and observability across service boundaries. Start with 2.4 FM API Integrations.

2.5 Application patterns and tools is about the developer and workflow layer: API Gateway, Step Functions, EventBridge, Prompt Flows, CI/CD, Amazon Q Developer, GenAI gateways, and business-system enhancements. Start with 2.5 Application Integration Patterns and Development Tools.

Integration decision stack

Use this order when a scenario asks how to implement a GenAI capability in a production application.

  1. Name the interaction pattern. Decide whether the app needs direct FM invocation, RAG, an agent, deterministic workflow orchestration, event-driven integration, batch processing, or a business-system enhancement.
  2. Constrain the authority boundary. Identify which tools, data sources, APIs, identities, and actions the FM or agent may access. AIP-C01 often rewards least privilege, parameter validation, human approval, and scoped tool definitions.
  3. Choose the runtime path. Match the workload to on-demand Bedrock invocation, provisioned throughput, SageMaker AI, Lambda, containers, API Gateway, queues, or hybrid deployment. Do not treat one serving option as universally correct.
  4. Define the invocation contract. Decide how the caller handles streaming, asynchronous work, retries, fallback, timeouts, schemas, validation, rate limits, and idempotency.
  5. Make it observable and releasable. Tie integration behavior to logs, traces, metrics, deployment pipelines, rollback criteria, and versioned prompts or workflows.

Work this domain in order

Start with 2.1 Agentic AI Solutions and Tool Integrations to lock down memory, state, tool boundaries, MCP, and approval-aware orchestration.

Then move to 2.2 Model Deployment Strategies for deployment targets, endpoint strategy, scaling, and production runtime choices.

Next review 2.3 Enterprise Integration Architectures for CRM, contact center, document, search, and business-system integration patterns.

Then cover 2.4 FM API Integrations for streaming, retries, fallback, structured outputs, routing, and invocation contracts.

Finish with 2.5 Application Integration Patterns and Development Tools for Step Functions, EventBridge, prompt flows, AppConfig, CI/CD, and operational development discipline.

Fast routing inside this chapter

  • Go to 2.1 Agentic AI Solutions and Tool Integrations when the scenario is about memory, session state, tool choice, approvals, MCP, multi-agent coordination, human review, or agent reasoning flow.
  • Go to 2.2 Model Deployment Strategies when the scenario is about deployment target, endpoint strategy, runtime scaling, availability, model cascading, hybrid hosting, or how the model-serving path should run.
  • Go to 2.3 Enterprise Integration Architectures when the scenario is about CRM, contact center, knowledge systems, identity federation, enterprise apps, business workflows, compliance-constrained routing, or cross-system architecture.
  • Go to 2.4 FM API Integrations when the scenario is about streaming, retries, asynchronous processing, structured outputs, response handling, model routing, request validation, rate limiting, or FM invocation APIs.
  • Go to 2.5 Application Integration Patterns and Development Tools when the scenario is about event-driven integration, workflow orchestration, Prompt Flows, GenAI gateways, developer productivity, troubleshooting tools, SDKs, or CI/CD patterns.

High-yield decision rules

  • If the workflow has known steps, prefer deterministic orchestration with Step Functions or event patterns before choosing a fully autonomous agent.
  • If the model must take action, inspect tool permissions, parameter validation, idempotency, error handling, human approval, and audit logging before discussing model capability.
  • If users need immediate feedback, evaluate streaming and client delivery. If work can happen later, evaluate asynchronous queues, callbacks, and batch patterns.
  • If the integration touches enterprise systems, identity federation, least privilege, API boundaries, observability, and rollback matter as much as prompt design.
  • If multiple models could answer, route by task complexity, latency, cost, quality requirement, and fallback behavior instead of picking one large model for everything.
  • If the app spans old and new systems, loose coupling through APIs, events, gateways, and synchronization patterns is usually safer than tight direct coupling.

Common traps

  • picking agent frameworks where deterministic orchestration is enough
  • skipping deployment/runtime thinking and treating Bedrock invocation as the whole architecture
  • calling tools without strong permission or approval boundaries
  • ignoring streaming, retry, idempotency, and fallback behavior
  • wiring GenAI into enterprise systems without clean auth, event, or audit design
  • treating API Gateway, Lambda, Step Functions, and EventBridge as interchangeable glue instead of choosing by request shape and failure mode
  • adding a GenAI gateway without defining policy, observability, routing, and ownership
  • building a custom integration when a managed service or existing business-system connector would reduce operational risk

In this section

Revised on Monday, June 15, 2026
1. FM Integration, Data, and Compliance
3. Safety, Security, and Governance