Make the intent explicit
Break requirements into tasks with acceptance criteria. Give each agent a clear role and a limited set of functions.
CHAPTER 04 / BEHIND THE PROJECTLearn the system.
Learn the system.
Own the decisions.
Cloud Learning Studio explains CSCE 412 Project 5 through a secure static website built around S3, CloudFront, Route 53, and ACM. Learn how bounded agents and human decisions shape the work.
THE ENGINEERING APPROACHSpecialized agents.
Specialized agents.
Bounded authority.
Five executable Python agents divide the work: Requirements, Website, Infrastructure, Security, and Verification/recovery. They exchange structured JSON and use role-specific tools.
The model interprets requirements, develops content, reviews evidence, and proposes actions. Deterministic Python validates arguments, calls approved APIs, checks actual results, and enforces human approval for cloud changes.
AI assistance is disclosed in the project documentation. Generated code and conclusions still require human understanding and independent validation.
A workflow you can explain.
REQUIREMENTS → ACTION → EVIDENCEKeep authority outside the model
Review the exact cloud plan. Approve its hash. Reject unrecognized tools and changed recovery policies.
Save what really happened
Capture structured exchanges, tool results, validation failures, and recovery evidence. Never turn a simulated result into a deployment claim.