AWS re:Invent 2025 - AI Native Development: Strategies and Impact across Amazon and AWS (DEV323)

AI Native Development: Strategies and Impact at Amazon and AWS

Transitioning from AI-Assisted to AI-Native Development

• Presenters shared lessons, challenges, and breakthroughs in moving from isolated AI experiments to AI-native transformation at Prime Video and AWS.
• AI-assisted development uses AI tools in an ad-hoc, isolated manner, with limited impact on individual productivity.
• AI-native development fundamentally reimagines the software development lifecycle, embedding AI capabilities across all roles and phases.

Key Pillars of AI-Native Transformation

1. Access and Infrastructure:

   • Providing cross-functional teams (PMs, designers, developers) with access to best-in-class AI tools and shared context via Model Context Protocol (MCP).
   • Leveraging AWS services like Bedrock, SageMaker, and Agent Core to build the AI infrastructure.
   • Balancing centralized and decentralized investments to enable both common and domain-specific capabilities.

2. Culture and Learning:

   • Fostering a grassroots "AI native learning flywheel" of experimentation, sharing, and healthy skepticism.
   • Formalizing mechanisms like AI champions, bar-raising programs, and quick-start guides to scale adoption.
   • Designing role-specific learning paths to upskill PMs, designers, and other stakeholders on AI-native techniques.

3. Trust and Rigor:

   • Addressing early challenges like code review overload and policy/process friction.
   • Embedding governance and controls at every layer of the AI infrastructure stack.
   • Enforcing permissions, capacity controls, and rigorous model evaluation to ensure safe experimentation and adoption.

Measuring the Impact of AI-Native Transformation

• Tracking adoption metrics across development lifecycle phases and roles.
• Monitoring velocity and efficiency improvements using industry standard DORA metrics, while also controlling for incident rates.
• Diving into specific use case metrics like time/effort savings for PM workflows.
• Gathering qualitative feedback through surveys, interviews, and leadership listening sessions.

Demonstration: Strands Agent SOPs

• Introduced "Strands Agent SOPs" - a new open-source framework for authoring and executing reusable, standardized AI agent workflows.
• Demonstrated how the SOP format allows agents to follow detailed, constrained steps to accomplish tasks like triaging GitHub issues.
• Showed how the SOP framework enables easy authoring and customization of agent workflows without having to manage low-level prompts.

Key Takeaways

• Successful AI-native transformation requires a balanced approach of grassroots energy and top-down organizational support.
• Providing access, building culture, and ensuring trust/rigor are critical pillars for scaling AI adoption across an enterprise.
• Measuring impact holistically, across productivity metrics and qualitative feedback, is essential for driving continuous improvement.
• The Strands Agent SOP framework offers a new paradigm for encapsulating and reusing AI agent workflows in a robust, transparent manner.