DEV Community

Khadija Asim
Khadija Asim

Posted on

Build an AWS Cost Optimization Agent That Ships to Production

#ai #hiring #startup

AWS cost dashboards are where recommendations go to die. FinOps tools routinely generate long lists of idle NAT Gateways, unattached EBS volumes, and overprovisioned ECS tasks. Yet, cloud bills remain high because reviewing dashboard recommendations requires manual context switching.
To fix this, cost optimization agents must act inside the existing developer workflow. Instead of generating static reports, an in-workflow agent detects waste, maps it to infrastructure code, and submits actionable pull requests.

Architecture of an In-Workflow Cost Agent

A production cost optimization agent relies on three core components:

  • Ingestion: Listens to AWS Cost Explorer anomalies, EventBridge events, or scheduled AWS SDK queries for unattached resources.
  • Context Mapping: Queries tag metadata and maps AWS Resource Names (ARNs) to Terraform or CloudFormation state files in your repository.
  • Remediation Payload: Generates a git diff or interactive Slack message with a precise action plan. Here is a simplified Python example using boto3 to identify unattached EBS volumes and construct a payload for a workflow trigger: 
import boto3
def check_unattached_ebs():
    ec2 = boto3.client('ec2')
    volumes = ec2.describe_volumes(
        Filters=[{'Name': 'status', 'Values': ['available']}]
    )
    remediations = []
    for vol in volumes['Volumes']:
        remediations.append({
            "volume_id": vol['VolumeId'],
            "size_gb": vol['Size'],
            "monthly_cost_usd": vol['Size'] * 0.08
        })
    return remediations
Enter fullscreen mode Exit fullscreen mode

Moving from Demos to Production Safety

Most AI agents stay stuck as internal demos because teams fear automated deletion of production infrastructure. To make an agent production ready:

  • Enforce Read-Only Scopes First: The agent analyzes resources and drafts Terraform PRs. Humans review and merge.
  • Add Workflow Approval Gates: For immediate actions, such as stopping non-production RDS instances on weekends, route execution through Slack interactive buttons backed by signed IAM roles.
  • Trace Every Action: Log all proposed and executed changes to immutable audit trails. Where agents pay for themselves is in reclaiming developer hours from routine operational debt. Most teams get a demo, but you need production. Engineering providers like https://gaper.io help teams build and deploy custom agents directly into production pipelines. For one client, pairing developer talent with a custom AI agent handling ticket triage cut manual support workload by an estimated 40%. The same in-workflow strategy applies directly to cloud infrastructure management. ## Conclusion Stop sending developers to third-party cost portals. Ship an agent that integrates into GitHub, reads your IaC repositories, and opens PRs to clean up unused AWS resources.

Top comments (0)