How to manage large complex projects
This walkthrough demonstrates how DuploCloud's AI DevOps suite takes an existing e-commerce order service running on AWS EKS from zero observability and security hardening to a fully production-ready, SOC 2-compliant platform — in a single automated workflow.
The Starting Point
A simple order service is deployed in a sandbox environment. The goal: scale the platform and pass a SOC 2 audit in 90 days. The microservice architecture runs 5–10 services on AWS EKS with no observability, no cost controls, no automated pipelines, and no security hardening.
Step 1 — Connect the Service to DuploCloud
Before any work can begin, the service must be connected to DuploCloud. Providers and scopes for both AWS and Kubernetes have already been created.
Verify AWS Connection — An AWS provider is configured with the correct credentials and scope. Confirming the sandbox nodes are running verifies DuploCloud is connected to the AWS account.
Verify Kubernetes Connection — A Kubernetes provider is configured with credentials and a scope defining what DuploCloud can access. Confirming the pods from the order service are visible verifies the connection.
Step 2 — Set Up the Workspace
All work inside DuploCloud happens inside workspaces. Select the workspace for this project — it already has the required skills and agents configured.
Scroll down to Cloud Providers and add the new scope for the service. Then under Kubernetes, add the scope that gives the system access to this particular service. Hit Update to save.
Step 3 — Create a Project
Navigate to AI DevOps → Projects. Create a new project, give it a name, and select the right scopes and skills. DuploCloud uses a spec-driven approach (as recommended by Anthropic) — open the AI Planner and provide a prompt with all project requirements.
DuploCloud runs a discovery process, asking for user input where required.
Once complete, it produces a spec document with high-level requirements, then a plan document with the step-by-step process divided into phases and tasks, and finally a list of execution tasks divided into stages.
Tip: Save the spec, plan, and execution task files — Claude Code can use them as context when picking up tasks directly in your IDE.
Step 4 — Project Dashboard and Tickets
Back on the project dashboard, the agent has created individual tickets for each task, grouped by phase.
Step 5 — Execute the First Task
Open the first ticket to find the current state — the metrics server is not installed and no metrics are reporting. DuploCloud works autonomously, installs the metrics server, and metrics are now being reported.
Step 6 — Collaborate via Claude Code
The next task is assigned to a team member (John) who prefers working from his IDE using DuploCloud's Claude Code plugin. He initializes the workspace and project and picks up right where the team left off.
John executes two tickets via Terraform — enabling EKS control plane logging to CloudWatch, then enabling Container Insights to collect pod-level metrics via a CloudWatch agent daemon set.
Everything done in Claude Code is automatically mirrored back in the ticket.
One Task at a Time, End to End
DuploCloud doesn't hand you a list of recommendations — it works through them. Each phase of the plan is broken into discrete tickets, and the agent executes them one by one: running commands, validating results, and moving to the next. Whether a task is handled autonomously or picked up by a team member via Claude Code, every action is logged and every outcome is verified before the project moves forward.
Last updated
Was this helpful?