Engineering notes / AWS platform operations
Practical explanations of GitOps, Amazon EKS, AI-assisted operations, and the engineering controls that make automation trustworthy.
Read the latest note
Featured engineering noteEngineering field notes
Practical architecture for CTOs, platform leaders, and engineers who need the operating decision and the implementation detail to make sense together.
A read-only Lambda ran about 5,760 times across one weekend. Its code looked simple, while the billable behaviour depended on log volume, access scope and API usage elsewhere in the platform.
Give AI the GitOps repository, IaC repository and read-only AWS and Kubernetes views. It can trace a change, expose missing dependencies and prepare linked pull requests for engineers to review.
An AI agent can follow its task correctly while acting on a wrong understanding of the environment. Engineers need a live view of its work and a stop mechanism outside the agent.
AI monitoring works best when the agent knows what healthy looks like, how services connect and what changed. A short report gives engineers a useful place to start.
An agent can understand each tool separately and miss how Git, CI, Argo CD, AWS, controllers and service dependencies affect one another. It needs the path from the original request to recovery.
Give each change the smallest environment needed for a confident review, and keep shared staging for work that genuinely needs a common state.
One commit can describe the full source state in a monorepo. The preview still needs clear rules for builds, data, dependencies, cost and cleanup.
A pull request in one repository describes one service. A useful preview must also choose known versions for every service around it and record them in one place.
We have seen tidy CIDR plans become a limit once EKS pods, load balancers, endpoints, peering and new environments start consuming addresses.
Installing Argo CD is usually straightforward. The harder work is deciding repository ownership, permissions, promotion, deletion, recovery and what a healthy result means.
Bring the real constraint
We connect the explanation, architecture, permissions, repositories, reconciliation, and operating evidence so the automation can be understood and owned by your team.
Schedule a platform call