AWS re:Invent 2025 - Designing with Agents: A Playbook for Enterprise Engineering Leaders (DVT102)

Designing with Agents: A Playbook for Enterprise Engineering Leaders

Introduction to Cognition and AI-Powered Software Engineering

• Cognition is an AI software engineering company founded by a team of competitive programmers and AI researchers
• They have developed advanced AI agents and tools to augment and automate software engineering tasks
• Cognition has seen significant growth, reaching over 200 employees and a $10 billion valuation

Three Waves of AI in Software Engineering

1. Co-Pilot Tab Completion: Real-time AI assistance for code completion and suggestions within IDEs (e.g. GitHub Copilot)
2. Full AI Development Environments: AI-powered IDEs that leverage AI for code understanding, documentation, and collaboration (e.g. Cognition's Windsurf)
3. AI Software Engineers: Autonomous AI agents that can handle entire software engineering tasks and workflows end-to-end

Patterns for Successful Enterprise Adoption of AI-Powered Engineering

1. Leveraging Multiple Modes of AI Interaction

• Synchronous, real-time AI assistance (e.g. Windsurf) keeps engineers in flow state
• Asynchronous, autonomous AI agents (e.g. Cognition's Devon) can handle larger, more complex tasks

2. Understanding Existing Codebase Context is Critical

• Cognition's DeepWiki product provides a comprehensive, searchable index of codebase context and architecture
• This enables AI agents to better understand and work within existing systems and requirements

3. Optimizing the Entire Software Development Lifecycle

• AI can automate not just code writing, but also testing, code review, security scanning, and other SDLC tasks
• Measuring and optimizing end-to-end project velocity, not just individual coding productivity

Deployment and Change Management Best Practices

• Recognize that AI-powered engineering is a workflow transformation, not just a new tool
• Combine top-down mandates with bottom-up buy-in and celebration of early successes
• Ensure AI agents have full access to the same tools and systems as human engineers
• Invest in training engineers on effective prompting and delegation to AI agents
• Measure impact at multiple levels, focusing on end-to-end business outcomes

Real-World Case Study: Largest Bank in Latin America

• Indexed over 300,000 repos with DeepWiki for improved context understanding
• 75% of 17,000 engineers using Devon AI agent in production
• 70% of security vulnerabilities automatically remediated by Devon
• 5-6x faster modernization and migration projects
• 2x increase in test coverage on critical systems

Key Takeaways

• AI-powered software engineering can drive transformative productivity gains, but requires a holistic approach
• Synchronous and asynchronous AI collaboration modes are both important, with clear delineation of use cases
• Comprehensive codebase understanding is a critical foundation for effective AI agents
• Automating the entire SDLC, not just coding, unlocks the biggest productivity improvements
• Successful enterprise adoption requires both technical and organizational change management