AI & ML Research 3 Days

Last updated: July 19, 2026, 11:30 AM ET

AI Cost and ROI Measurement

OpenAI's CFO, Sarah Friar, introduced a practical AI scorecard measuring ROI through metrics like useful work, cost per successful task, dependability, and return on compute. This comes as an AI agent demonstrated success in evaluations but was ultimately deemed too expensive by its CFO, costing more than the humans it replaced. This highlights a critical challenge in AI adoption: ensuring that successful task completion translates into economic viability.

Building AI-Native Platforms and Architectures

Many organizations are implementing AI but struggle to construct an AI-native enterprise data platform. A practical architecture involves data agents, AI-powered QA, and robust AI governance. Engineers are exploring "loop engineering," with some experiments focusing on architectures that function without an LLM at the center. Furthermore, classical Machine Learning techniques are being leveraged to empower AI agents, building on existing foundations for enhanced capabilities.

Optimizing LLM Interactions and Document Processing

Effective utilization of advanced models like GPT-5.6 requires specific strategies to maximize their potential in various applications. In enterprise document intelligence, a cost-effective approach involves "loop engineering" with adaptive PDF parsing, where cheaper, deterministic checks are performed first, escalating to more expensive parsers only when necessary. A production RAG pipeline is demonstrated, showing how to wire together components for direct citation of answers.

Preparing for Increased AI Agent Workloads

As AI agents are tasked with more responsibilities, organizations must prepare by defining recurring work, providing agents with the right context, and clearly communicating expectations for high-quality output. This includes identifying areas where human judgment remains essential. The challenge of balancing AI efficiency with human oversight is further emphasized by the fact that even agents passing all technical evaluations can be economically unviable.

Emerging Trends and Safety in AI

Analog AI is experiencing a resurgence, driven by the energy demands of current AI systems, with a focus on using physics for computation instead of traditional digital logic. In parallel, efforts are underway to ensure AI safety, particularly for younger users. OpenAI is implementing age-appropriate protections, learning tools, and parental controls to make Chat GPT safer for teens. Separately, concerns are rising about the potential for weather data sabotage, which could impact critical decisions made by industries including aviation, energy, and agriculture.