AWS re:Invent 2025 - Introducing AI driven development lifecycle (AI-DLC) (DVT214)

Introducing AI-Driven Development Lifecycle (AI-DLC)

Overview

• Presenters: Anupam Mishra, Director of Solution Architecture at AWS, and Raja, who leads the Developer Transformation team at AWS
• Goal: Share learnings and best practices for using AI in software development

Challenges with Current AI Adoption

• Two common approaches:
  1. AI-Managed Approach: Developers throw complex problems to AI, expecting it to solve end-to-end autonomously. This rarely works except for simple prototypes.
  2. AI-Assisted Approach: Developers take tight control, using AI only for narrow tasks. This leads to slow progress and limited productivity gains.
• Key issues:
  • AI makes many assumptions and produces code that developers lack confidence in
  • Existing processes and workflows are not optimized for the speed of AI-powered development

The AI-DLC Methodology

• Principles:
  1. Align AI's "brain" with the human's through iterative planning, validation, and execution
  2. Redesign workflows to enable rapid, synchronized collaboration between humans and AI
• Key Rituals:
  1. Mobile Elaboration: Product managers, developers, QA, and ops collaborate to rapidly refine requirements and define stories using AI
  2. Mob Construction: Cross-functional teams work together in the same physical or virtual space, using AI to generate code quickly
• Adaptive Workflow:
  • Consists of 9 stages across Inception, Construction, and Operation phases
  • Stages are customized based on the complexity of the problem (e.g., a simple bug fix vs. a new feature)
  • AI recommends the appropriate stages, which are then validated and adjusted by the team

Technical Learnings

1. Understanding AI Capabilities: AI excels at generating code but struggles with understanding context, dependencies, and overall system design. Developers must remain involved to provide oversight and guidance.
2. Managing Context Windows: Keeping the context narrow and focused is crucial to avoid confusing the AI.
3. Mimicking Existing Code: AI performs best when given reference code to use as a template, rather than describing the desired functionality.
4. Increasing Release Velocity: Rapid development with AI requires comprehensive unit and integration testing to maintain quality.
5. Accounting for Model Biases: AI models are trained on existing code, so they may struggle with newer languages or design patterns not well-represented in the training data.
6. Maintaining Flow State: Continuous, uninterrupted work sessions are essential for developers to stay in sync with the AI.
7. Investing in DevOps Maturity: Robust CI/CD pipelines and well-oiled development environments are critical to support the increased pace of AI-powered development.

Business Impact and Examples

• Customers have reported 3-10x productivity gains using the AI-DLC methodology, with improved quality and predictability.
• Examples:
  • System integrator Vipro completed 3 months of work in 20 hours using AI-DLC.
  • Fintech company Dun built a new application in 48 hours and launched it the following week.

Getting Started with AI-DLC

• AWS provides self-help resources, workshops, and an open-source workflow to help organizations adopt the AI-DLC methodology.
• The presenters also offer a "Unicorn Gym" program where AWS engineers collaborate with customers to solve real-world problems using AI-DLC.
• An "AI-Native Builders Community" is being established to collect and share learnings, and define the future of AI-powered software development.