Build and deploy AI agents on Google Cloud

This skill guides agents through the workflow of designing and implementing a tailored multi-product solution in the cloud for a given workload, use case, or requirement.

Workflow

The solution design and implementation workflow is divided into the following phases:

• Phase 1: Requirements discovery and analysis: Analyze the workload's requirements, constraints, dependencies, and current state.
• Phase 2: Solution design: Build a technology stack, architecture, and deployment configuration for the workload based on Google Cloud design best practices and recommendations.
• Phase 3: Implementation plan: Generate automation and instructions to deploy the solution.
• Phase 4: Solution validation: Validate that the deployment meets the requirements of the workload.

Copy this checklist into your active task/plan artifact to track progress across the four phases:

• Phase 1: Requirements discovery and analysis completed & confirmed.
• Phase 2: Solution architecture generated & approved.
• Phase 3: Implementation plan generated & approved.
• Phase 4: Solution validation generated & approved.

Phase 1: Requirements discovery and analysis

1. Discover requirements: Gather and understand the functional and non-functional requirements, business goals, and current state (if any) of the workload, including its architecture, dependencies, and constraints.

   Important: First, check whether the user's initial prompt has already answered the following questions or whether the prompt explicitly asks you to propose a solution architecture/diagram from a given set of parameters.

   • If the user's prompt provides sufficient requirements and it explicitly requests an architecture proposal or diagram, then skip asking the questions below, and instead proceed to the step Recommend agent design pattern.

   • If the user's prompt doesn't provide sufficient requirements, then complete these steps to gather missing information:

     1. Ask the user to describe the functional requirements of their workload: business processes, activities, and use cases.

     2. Ask the user to describe the non-functional requirements (security, privacy, compliance, reliability, disaster recovery, cost, operations, performance, and sustainability) of their workloads.

     3. Ask the user what existing systems, knowledge bases, product documentation, or other documentation the AI agents need to access for grounded guidance.

     4. Ask the user to describe dependencies, if any, on other workloads, products, or tools.

     5. Review the input that the user has provided so far, and check whether there are any ambiguities or contradictions in the input.

        If you identify any ambiguities or contradictions in the requirements that the user has provided, then do the following for each ambiguity or contradiction that you identify:

        • Describe the ambiguity or contradiction.
        • Ask the user how they wish to resolve the ambiguity or contradiction.
          • If the user delegates the choice to you (e.g., the user replies with "do what you think is best" or "you decide"), then provide a clear suggestion to resolve the ambiguity or contradiction, explain your reasoning, and ask the user to approve your suggestion.

        Critical: Until all the ambiguities and contradictions that you identify are resolved according to the preceding guidance, you must NOT recommend or generate any architecture design, technical decomposition, or Google Cloud product recommendations.

2. Recommend agent design pattern: Evaluate the complexity, workflow, latency, and cost requirements of the workload to recommend an agent design pattern:

   • Single-agent system: Recommend for simpler tasks, acting as an effective starting point to refine core logic and tools.
   • Multi-agent system: Recommend for complex problems requiring multiple specialized agents to collaborate on a workflow.

3. Identify components: Based on the requirements analysis, generate a technical decomposition of the workload. The technical decomposition must identify the logical components of the workloads and their relationships. Also identify any cross-cloud components, hybrid components, or on-premises components that the solution needs to integrate with.

4. Ask for confirmation: Ask the user to confirm whether the recommended design pattern and technical decomposition match their workload requirements.

5. Iterate: If the user requests changes, generate an updated technical decomposition, and ask the user to confirm the changes. Continue iterating until the user confirms the technical decomposition. Proceed to the next phase only after the user provides confirmation of the technical decomposition.

Phase 2: Solution design

1. Retrieve relevant Google Cloud guidance from references/related-guidance.md.

   Important: Use the content that you retrieved from references/related-guidance.md to ground the guidance that you generate in the remaining steps of this phase.

2. Map components to Google Cloud products: For each component in the confirmed technical decomposition, identify the appropriate Google Cloud products and features by consulting product-mappings.md for detailed recommendations, trade-offs, and alternatives across networking, frontends, agent/model runtimes, memory stores, and tools.

3. Create architecture diagram: Create an architecture diagram in Mermaid format: https://github.com/mermaid-js/mermaid. The diagram should show the components, their relationships, and data/control flows.

4. Generate design recommendations: Generate design guidance based on the following Google Cloud best practices and recommendations. Use the information in references/related-guidance.md, with an emphasis on the guidance in references/design-principles.md.

5. Draft solution architecture: Compile the requirements, technical decomposition, product mapping, architecture diagram, and design recommendations into a single Markdown file adhering to the format in solution-template.md. Save this document in the workspace as solution-architecture.md.

6. Request review: Present the generated solution architecture (including the complete fenced mermaid code block for the diagram) directly to the user in your response, and explicitly request their feedback or approval. When you present the architecture, ask the user to provide approval for you to proceed with an implementation plan.

7. Iterate: If the user requests changes, generate an updated solution architecture and repeat the steps from "Map components to Google Cloud products" through "Request review" until the user approves the solution architecture.

Phase 3: Implementation plan

1. Retrieve relevant implementation resources:

   Important: Use the resources in references/related-guidance.md as the technical foundation for the Infrastructure as Code (IaC) and the deployment instructions that you generate in the remaining steps of this phase.

2. Identify deployment prerequisites: Document prerequisites for the deployment, including the following:

   • Projects and billing associations
   • Required Google Cloud APIs
   • Required IAM permissions
   • Any other prerequisites

3. Generate Infrastructure as Code (IaC): Generate code (e.g., Terraform) and deployment scripts to automate the provisioning of the proposed Google Cloud resources.

   • Where appropriate, alongside or instead of raw infrastructure scripts, instruct the user to use Agents CLI commands (agents-cli scaffold create or agents-cli scaffold enhance) to set up or enhance the project structure, deployment configuration, and CI/CD pipelines.

4. Write deployment instructions: Draft sequential, step-by-step deployment instructions to execute the IaC and initialize the workload components. Compile the deployment prerequisites, IaC, and deployment instructions into a single Markdown file adhering to the format in implementation-template.md. Save this document in the workspace as implementation-instructions.md.

   • The instructions MUST provide the exact ADK code to define a stateful agent node that takes a prompt, calls a model, and returns a tool execution request.
   • The instructions MUST demonstrate how to register tools like database readers by using Model Context Protocol (MCP) standards.
   • If deploying the agent to Cloud Run, the instructions MUST show how to configure Cloud Run to scale to zero when the agent is idle, reducing runtime costs.
   • The instructions MUST recommend using encrypted environment variables to store model parameters or private API credentials. Encryption helps to prevent the exposure of sensitive credentials in plain-text container log streams.
   • Where appropriate, the instructions MUST specify using the Agents CLI agents-cli deploy command (alongside or instead of raw infrastructure/deployment scripts) to run the deployment.

5. Request review: Present the generated deployment instructions to the user and explicitly request their feedback and confirmation.

6. Iterate: If the user requests changes, generate an updated implementation plan and repeat the steps from "Generate Infrastructure as Code (IaC)" through "Request review" until the user approves the implementation plan.

Phase 4: Solution validation

1. Retrieve relevant verification resources:

   Important: Use the resources in references/related-guidance.md and their verification patterns as the starting point for the validation checks and verification scripts that you generate in the remaining steps of this phase.

2. Define validation checks: Outline validation steps to verify that the deployed infrastructure meets the workload's requirements:

   • Deployment dry-run: Commands like terraform plan to preview changes. Include instructions to run agent deployment in dry-run mode (e.g., using agents-cli deploy --dry-run or -n) to preview steps and Terraform executions before pushing to production.
   • Local testing and quality verification: Recommend using the Agents CLI to run and test agent logic locally (agents-cli run) and conduct systematic evaluations (agents-cli eval run) to verify agent quality and performance before deploying.
   • Connectivity and routing: Verification of network paths, load balancer routing, and service endpoints.
   • Security policies: Verification of restricted access, firewall rules, and IAM enforcement.

3. Generate verification scripts: Draft lightweight scripts or command-line instructions (e.g. using curl, gcloud, or agents-cli) that the user can run to perform these validation checks.

   • The validation plan MUST include instructions using the Agents CLI for local runs, evaluations, and post-deployment validation checks (e.g., agents-cli run --url <service-url> to test the deployed service endpoint).

4. Compile validation plan: Document the validation steps, verification scripts, and expected outcomes in a single Markdown file adhering to the format in validation-template.md. Save this document in the workspace as validation-plan.md.

5. Request review: Present the validation plan to the user and explicitly request their feedback or approval on the validation plan.

6. Conduct validation and finalize: Assist the user in executing the validation checks and troubleshooting any deployment issues. After the solution is validated successfully, request final approval from the user.

7. Iterate: If the user requests changes, generate an updated validation plan and repeat the steps from "Define validation checks" through "Request review" until the user approves the validation plan.

References & Supporting Links

• For the complete list of Google Cloud architectural documentation, product manuals, development kits, and checklists used by this skill, see related-guidance.md.