← back

AI-Driven Life Cycle Development (AI-DLC)

When coding with agents I’ve found myself losing track of changes to my application quite frequently. I’d start with building a feature -> ask the agent to add another feature -> modify the first feature -> add another feature -> a previous feature breaks -> I ask the agent to fix it -> and now I’ve lost track of how my application even works

But I think I might’ve found a fix for this now with AI-driven life cycle development (AI-DLC). It follows the personas of traditional SDLC but it removes the bits no longer needed when doing AI driven development. The idea is to have a all the personas talk to each other and follow the steps of traditional SDLC through prompts. We would be planning and documenting the entire process to keep track of things.

Inception -> Construction -> Operation AI Governance underlying all of it

Here is an example when building a Greenfiield version (MVP):

  • Phase 1: Ground Rules & Intent: Set strict boundaries. The AI must create a markdown plan and wait for my approval before executing anything. Defined the core MVP intent.
  • Phase 2: Product Management: Assigned a “Product Manager” persona to generate formal user stories and acceptance criteria from the core requirement.
  • Phase 3: UX/UI Design: A “UX/UI Designer” persona translated the user stories into visual wireframe descriptions and a frontend component hierarchy.
  • Phase 4: Architecture & Data Design: Using “Software Architect” and “Database Architect” personas, the AI mapped out the system components, API contracts, and the database schema/migrations.
  • Phase 5: Full-Stack Development: A “Software Engineer” persona generated the frontend UI components and wrapped the core backend logic into functional API endpoints.
  • Phase 6: Cloud Deployment: Acting as a “Cloud Architect,” the AI built an Infrastructure-as-Code deployment plan (e.g., CloudFormation) and strategized AI-driven production monitoring.
  • Phase 7: DevOps: A “DevOps Engineer” persona documented the local setup, configured CI/CD pipelines, initialized Git, and pushed the complete project to the repository.

I’ve already started using this approach for my development projects and it has made it easier to keep track of changes for me.

Here’s my GitHub repo for anyone curious to see an example workflow: https://github.com/ankittejyadav/task_manager/blob/main/aidlc-docs/prompts.md

#GenerativeAI #AWS #Claude #SoftwareEngineering #MachineLearning #MIT #AgenticAI #CloudArchitecture #TechCommunity

← all posts