# AI Devworld Conference — Agenda Web agenda: https://aidevworld.ai/agenda Plain-text agenda: https://aidevworld.ai/agenda.txt ## Day 1 ### The AI Product Engineering Playbook - Time: Icon: mic-vocal - Type: Keynote - Track: AI Product Engineering - Level: All - Topics: AI, Product The industry has moved beyond prototypes. Teams are now responsible for building production AI systems that create measurable business value. This keynote introduces the emerging discipline of AI Product Engineering and explores the architectural patterns, organizational structures, and engineering practices separating successful AI products from expensive experiments. ### AI-Native UX Patterns: Beyond Chat Interfaces - Time: Icon: mic-vocal - Type: Talk - Track: UX Patterns - Topics: AI, Python Streaming responses, inline copilots, confidence indicators, approval workflows, collaborative generation, progressive disclosure, and human override mechanisms are becoming foundational interaction patterns. This session explores the design primitives emerging across modern AI products and examines why traditional CRUD application thinking fails when systems become probabilistic and generative. ### Evaluation-Driven Development: The Missing Engineering Discipline - Time: Icon: mic-vocal - Type: Talk - Track: eval - Level: Advanced - Topics: React, JavaScript How do you know your RAG pipeline improved quality? How do you verify a prompt refactor didn’t regress performance? How do you safely ship a new model? This session introduces Evaluation-Driven Development (EDD), a systematic approach to building AI systems using golden datasets, task-level benchmarks, LLM-as-a-judge techniques, pairwise comparisons, synthetic evaluation generation, regression testing, and continuous quality monitoring. Attendees will leave with a practical framework for integrating evaluations into their development lifecycle. ### Structured Outputs, Function Calling & Type-Safe Generation - Time: Icon: mic-vocal - Type: Talk - Track: Structured Outputs - Level: All - Topics: AI, Leadership Production systems cannot depend on fragile string parsing. Learn how engineering teams leverage JSON schemas, structured outputs, constrained decoding, tool invocation contracts, validation layers, and strongly typed interfaces to build reliable AI-powered applications. We’ll examine patterns for integrating LLM outputs into TypeScript, Python, and enterprise systems without sacrificing robustness. ### Lightning Block (5x5 minutes) - Time: Icon: mic-vocal - Type: Lightning Talks - Track: Snack Tips - Level: Intermediate - Topics: TypeScript 1. Why Prompt Engineering Is Dead 2. The Rise of Context Engineering 3. Small Models, Big Results 4. Copilot UX Anti-Patterns 5. What We Learned from Failed RAG Systems ### RAG in Production: Retrieval Systems That Actually Work - Time: Icon: mic-vocal - Type: Talk - Track: Rag - Topics: AI, Product Most retrieval systems fail because teams underestimate retrieval quality and overestimate model capability. This session dives into chunking strategies, hybrid retrieval, BM25 + vector search, reranking pipelines, metadata filtering, contextual compression, semantic caching, and retrieval evaluation techniques. Learn the architectural tradeoffs behind high-performing production RAG systems. ### Context Engineering: The New Prompt Engineering - Time: Icon: mic-vocal - Type: Talk - Track: Context Engineering - Level: All - Topics: AI As foundation models become increasingly capable, context becomes the primary lever for improving system performance. This session explores context construction pipelines, memory hierarchies, personalization layers, retrieval augmentation, tool context injection, conversation state management, and dynamic context assembly. We’ll examine why many teams now view context engineering as the most important layer in the stack. ### Agent Experience (AX): The Missing Layer in AI Product Engineering - Time: Icon: mic-vocal - Type: Talk - Track: AX Most organizations are focused on building agents. Very few are asking a more important question: Are our systems designed to be used by them? As AI agents become capable of planning, reasoning, retrieving context, invoking tools, and executing workflows, traditional user-centric design approaches begin to break down. This session introduces Agent Experience (AX), a framework for designing software systems that are understandable, discoverable, observable, and operable by autonomous agents. We’ll explore MCP, tool schemas, capability discovery, structured outputs, context engineering, machine-readable workflows, agent permissions, and the architectural patterns emerging across agent-native platforms. Because the next generation of software won’t just serve human users. It will serve agents too. ### Human-in-the-Loop Architectures - Time: Icon: mic-vocal - Type: Talk - Track: Human in the Loops The most successful AI systems are not autonomous—they are collaborative. Learn how teams design review workflows, approval gates, escalation paths, exception handling, and operational controls that combine human judgment with machine efficiency. Real-world case studies will illustrate where automation ends and human oversight begins. ### Agentic Workflows vs Autonomous Agents - Time: Icon: mic-vocal - Type: Talk - Track: Agent vs Workflow Should you build an agent or a workflow? This session explores planner-executor patterns, state machines, orchestrators, deterministic workflows, reflection loops, task decomposition, and multi-step reasoning architectures. We’ll separate engineering reality from marketing hype and discuss when agents genuinely outperform traditional orchestration. ### MCP and the Emerging Tool Ecosystem - Time: Icon: mic-vocal - Type: Talk - Track: Model Context Protocol Model Context Protocol is rapidly becoming the interoperability layer between models and software systems. Learn how MCP servers, tool registries, capability discovery, permissions, and context providers are reshaping application architecture and enabling a new generation of AI-native software ecosystems. ### From Prototype to Production - Time: Icon: mic-vocal - Type: Talk - Track: Go to Market Moving from a successful demo to a production system introduces entirely new challenges. We’ll cover rollout strategies, model versioning, prompt management, evaluation pipelines, monitoring, feature flags, experimentation frameworks, and operational readiness reviews. ### Building AI Products for Millions of Users - Time: Icon: mic-vocal - Type: Keynote - Track: Case Study A deep technical case study examining architecture decisions, evaluation frameworks, infrastructure tradeoffs, organizational lessons, and operational realities behind a large-scale AI product deployment. ## Day 2 ### The Reality of Operating AI Systems - Time: Icon: mic-vocal - Type: Keynote - Track: Operating AI Systems The launch is the beginning. This keynote examines the operational maturity model for AI-powered systems and explores how teams evolve from experimentation to reliability, governance, platformization, and organizational scale. ### AI Observability: Tracing the Entire Inference Path - Time: Icon: mic-vocal - Type: Talk - Track: AI Observability Traditional observability stops at API calls. AI observability requires visibility into prompts, retrieval pipelines, context construction, tool execution, agent decisions, and model outputs. Learn how modern tracing systems help engineering teams debug, evaluate, and continuously improve production AI systems. ### Reliability Engineering for Probabilistic Systems - Time: Icon: mic-vocal - Type: Talk - Track: Reliability Engineering Traditional software either works or fails. AI systems introduce ambiguity, uncertainty, and degradation modes that require entirely new reliability strategies. This session covers fallback architectures, model routing, graceful degradation, confidence thresholds, retry strategies, and resilience patterns for production AI systems. ### AI Security: Prompt Injection, Tool Abuse & Data Exfiltration - Time: Icon: mic-vocal - Type: Talk - Track: Security Prompt injection has become the SQL injection of the AI era. Learn how attackers exploit context windows, manipulate tool calls, extract sensitive information, and bypass safeguards. We’ll explore threat models, mitigation strategies, isolation techniques, capability scoping, and practical defensive architectures. ### Lightning Block (5x5 minutes) - Time: Icon: mic-vocal - Type: Lightning Talks - Track: Snack Tips * Jailbreaks We Didn’t Expect * MCP Security Lessons * Agent Failure Stories * Context Window Hacks * Observability Wins ### Cost Engineering for AI Products - Time: Icon: mic-vocal - Type: Talk - Track: Cost Management Scaling AI products requires balancing latency, quality, and spend. This session explores model routing, semantic caching, speculative execution, retrieval optimization, batching strategies, token reduction techniques, and inference cost management. Learn how teams reduce costs without sacrificing user experience. ### Agent Evaluation Frameworks - Time: Icon: mic-vocal - Track: eval Agent quality is notoriously difficult to measure. This session covers task completion metrics, trajectory evaluation, simulation environments, synthetic benchmarks, agent scoring systems, reward models, and production feedback loops. Learn how engineering teams determine whether agents are actually improving. ### Building Internal AI Platforms - Time: Icon: mic-vocal - Type: Talk - Track: architecture As adoption grows, organizations increasingly centralize capabilities into shared platforms. We’ll examine architecture patterns for model gateways, evaluation services, prompt registries, vector infrastructure, policy enforcement, governance layers, and developer self-service tooling. ### Multi-Agent Systems: Architecture, Coordination & Failure Modes - Time: Icon: mic-vocal - Type: Talk - Track: Multi Agent What happens when multiple agents collaborate? This session explores coordination strategies, blackboard architectures, agent handoffs, planner-worker systems, communication protocols, and operational challenges emerging in multi-agent environments. ### Governance Without Killing Innovation - Time: Icon: mic-vocal - Type: Talk - Track: Governance How do organizations maintain velocity while managing risk? Learn how leading teams implement guardrails, policy enforcement, model governance, auditability, compliance controls, and deployment workflows without creating bottlenecks for engineering teams. ### Coding Agents and the Future of Software Engineering - Time: Icon: mic-vocal - Type: Talk - Track: Job Security Software development is becoming one of the first professions transformed by AI. This session explores autonomous coding agents, agent-assisted development workflows, repository reasoning, codebase memory, verification strategies, and the changing role of engineers in AI-native teams. ### AI Product Engineering in 2030 - Time: Icon: mic-vocal - Type: Keynote - Track: Future Proofing A forward-looking exploration of where AI Product Engineering is heading: ubiquitous agents, adaptive interfaces, personalized context layers, autonomous workflows, and the new responsibilities of engineers building the next generation of software.