The AI-native architecture: Building a 'body' for the LLM brain

Most "AI-powered" tools are thin wrappers around a prompt, leading to hallucinations and fragile workflows. This session explains that truly AI-native platforms require more than treating LLMs as standalone solutions. Mark Brocato, head of engineering for Tonic Fabricate at Tonic.ai, outlines the architectural shift needed for AI-native development. He details the critical scaffolding required for internal platforms, including secure sandboxes for executing AI-written code, anti-assumption protocols that prioritize system state over model prediction, and the integration of deterministic logic with probabilistic outputs. Attendees learn how to build robust agentic systems that go beyond the chatbot to address high-stakes engineering problems.