Engineering AI 25 Session 3

Think MCP servers are complex enterprise-grade infrastructure? Think again. Ben Tayor breaks down the surprisingly straightforward process of building Model Context Protocol servers that extend AI capabilities in meaningful ways—demonstrated through a real code sandbox implementation that lets LLMs safely execute code. You'll see exactly how Ben built an MCP server that connects AI assistants to a secure WebAssembly-based code execution environment, transforming conversational AI from a text generator into a fully capable programming partner. This isn't theoretical—you'll watch real examples of Claude solving LeetCode problems and calculating prime numbers by writing and running actual code. Ben will cover <ul> <li>The MCP Mental Model — Understand how the Model Context Protocol creates a standardized bridge between AI assistants and external tools, and why this matters more than just building another API.</li> <li>From Concept to Code — Follow the actual development process of a production MCP server, including the key architectural decisions, WebAssembly integration patterns, and security considerations for sandboxed execution.</li> <li>Real-World Integration — See live demos of AI assistants using the code execution server, including debugging sessions, algorithm implementation, and the kind of collaborative coding that feels like magic but is actually just good engineering.</li> <li>The Bigger Picture — Discover how MCP servers are reshaping AI workflows, enabling everything from database connections to specialized tool integrations, and why building your own might be easier than you think.</li> </ul>

Integrating AI capabilities into existing software systems fundamentally changes how we approach software engineering. This talk explores the engineering practices, architectural considerations, and operational challenges that development teams face when their tech stacks need to support AI-powered features. Drawing from real-world experiences at CultureAmp and Appear, this session will cover the practical engineering challenges of preparing existing systems for AI integration: <h4>Engineering Practice Evolution:</h4> <ul> <li>How Model Context Protocols (MCPs) change service integration patterns - their security models, access scoping, API design for LLM consumption, and documentation requirements</li> <li>Adapting traditional software engineering practices for non-deterministic systems</li> <li>New testing and evaluation frameworks needed for AI-integrated applications</li> <li>Code review processes when dealing with prompt engineering and model interactions</li> </ul> <h4>Operational Engineering Challenges:</h4> <ul> <li>Observability patterns for agentic systems - monitoring, logging, and debugging non-deterministic behaviors</li> <li>Durability and reliability engineering when dealing with external AI services</li> <li>Performance engineering considerations for latency-sensitive AI integrations</li> <li>Error handling and graceful degradation patterns</li> </ul> <h4>Architectural Impact:</h4> <ul> <li>How existing microservices architectures adapt to support AI workflows</li> <li>Data pipeline engineering for model feeding and result processing</li> <li>Security engineering practices for AI-integrated systems</li> </ul> This talk focuses on the engineering discipline changes rather than product development, examining how traditional software engineering practices evolve when AI becomes a core component of your technical architecture. Attendees will learn practical strategies for evolving their engineering practices and technical systems to successfully integrate AI capabilities.

The thought of many coding agents spinning up across a mature codebase fills some engineering managers with dread, worrying about how to manage all the changes, deal with potential merge conflicts, half complete PRs, and agents going into a spiral and simply building the wrong thing. However, if you suddenly onboarded twenty grads into your development team, you would see many similar problems because the underlying systems and processes don't support a sudden influx of new developers into the codebase. Coding agents require some specific guidance, but systems and mechanisms you implement to help scale your AI development agents can also provide additional support for your human developers when they are onboarding to a new codebase or a new part of the stack - making your dev team more resilient and productive in the long run too. In this session, we'll cover how to implement engineering systems and processes that benefit coding agents as well as humans and make the whole team more effective and productive and the codebase more sustainable.

Stop fighting with AI tools and start architecting with them. Most engineers are stuck in prompt-and-pray mode, but the real productivity gains come from systematic workflows that turn AI from a frustrating chatbot into a strategic partner. This talk reveals three battle-tested patterns that will fundamentally change how you build software. You'll see exactly how to compress months of learning into days, why your long ChatGPT conversations always derail (and how to fix it), and discover the "scratchpad" technique that prevents you from making the same mistakes twice. <ul> <li>The Iterative Workflow — Learn why well-formed problems beat clever prompts every time, and see the strategic framework for choosing between Cursor, Claude, and Gemini based on what you're actually trying to accomplish.</li> <li>Context Management — Master the art of keeping AI conversations productive with "Clean Slate" resets, "Summary Carry-over" techniques, and the game-changing "Scratchpad" method that maintains institutional memory across projects.</li> <li>The Onboarding Accelerator — Watch a real case study of mastering Google's Agent Development Kit in record time by pointing AI at codebases and asking the right architectural questions.</li> </ul> This isn't about productivity hacks—it's about becoming an architect of AI-augmented workflows that amplify your engineering expertise and accelerate your entire team's capability.