Arthur

No-code and low-code AI tools have made it possible for anyone in an organization to build and deploy AI-powered applications — including the C-suite. But when business users create AI apps without understanding security fundamentals or model behavior, enterprises face a new class of risk: exposed data, unreliable outputs, and apps operating entirely outside governance frameworks. In this session, Zach Fry (CTO, Arthur) and Daniel Bardenstein (CEO, Manifest Cyber) bring together two critical lenses on enterprise AI governance — application security and model observability — to unpack what's actually at stake and what leaders can do about it. From prompt injection attacks on customer chatbots, to poisoned open-weight models, to AI agents autonomously deleting files and granting refunds they shouldn't — the risks are real, they're happening now, and most enterprises aren't ready. What's covered: - Why shadow AI is deeper and more dangerous than most security teams realize - The hidden risks in open-weight models and AI supply chains - How agents are breaking traditional identity and access management - Build-time vs. runtime enforcement: where should you focus first? - Practical steps to jumpstart your AI governance program Whether you're in security, engineering leadership, or responsible for AI strategy, this is the conversation your organization needs to be having. Chapters: 00:00 Introduction & the AI governance gap 04:30 What can go wrong: supply chain attacks, shadow AI & real-world incidents 14:00 How agents are entering the enterprise (and why IT can't keep up) 22:00 The cost of inaction: financial, security & operational risk 31:00 What good looks like: inventory, provenance & open-weight model risk 40:00 Build-time vs. runtime enforcement 47:00 Q&A: where to start & who owns liability #EnterpriseAI #AIGovernance #AISecurity #AgentSecurity #CyberSecurity

Nori Tatsumi, Head of Platform Engineering at Arthur AI, demos how the Arthur team turned a vibe-coded Slack Jira bot into a reliable, production-ready AI agent using Arthur Engine's own tracing, continuous evaluations, and prompt management tools. In this talk from Future of DevEx NYC, Nori walks through the full journey: instrumenting the agent from day one with OpenTelemetry tracing, writing binary pass/fail evals before making any changes, iterating on prompts without redeploying code, and letting evals catch regressions automatically. If you're building AI agents and wondering how to go from prototype to production, this talk covers the practical workflow — the Agent Development Lifecycle (ADLC) — that Arthur's engineering team pioneered and recommends to customers. What you'll learn: → How tracing reveals what your agent is actually doing under the hood → Why you should write evals before fixing anything → How prompt management lets you iterate without code deploys → How continuous evals catch issues before your users do --- About Arthur AI: Arthur gives developers the tools they need to ship reliable AI agents — from observability and tracing to prompt management, continuous evals, experiments, and more. Learn more at https://www.arthur.ai #aiagents #continuousimprovement #agentdevelopment #vibecoding #llm #ai