Inside OpenAI Enterprise: Forward Deployed Engineering, GPT-5, and More | BG2 Guest Interview

Quick Overview

OpenAI's enterprise platform, led by Sherwin Wu and Olivier Godement, focuses on distributing AGI benefits through its API and direct enterprise solutions, demonstrating significant success with customers like T-Mobile and Amgen, while also developing advanced models like GPT-5 and pioneering real-time voice and reinforcement fine-tuning (RFT) for customized, high-impact AI deployments.

Key Points: OpenAI's enterprise strategy centers on its API and direct enterprise products, viewing them as crucial for distributing AGI benefits widely, as stated by Olivier Godement: "we actually view our platform and especially our API and how we work with our customers, our enterprise customers, as our way of getting the benefits of AGI, of AI, to as many people as possible." Successful enterprise deployments include T-Mobile, where OpenAI models automate text and voice customer support, and Amgen, where AI accelerates drug development and speeds up regulatory processes, with Amgen being a "top customer of GPT-5, for instance." A unique deployment involved Los Alamos National Labs, requiring a custom on-premise setup on a supercomputer for sensitive national security research, showcasing flexibility beyond standard API offerings. GPT-5, described as "amazingly intelligent," offers significant improvements in reasoning, coding, and reduced hallucinations, though a trade-off exists between reasoning depth and latency, with customers sometimes opting for quicker, less optimal answers. OpenAI has launched a real-time API for voice, integrating speech-to-text, reasoning, and text-to-speech into a single, natural-sounding experience, moving beyond the 'stitch model' for improved latency and signal preservation. Model customization is deeply invested in, with Reinforcement Fine-Tuning (RFT) emerging as a powerful method for customers to create best-in-class models for their specific use cases by leveraging custom data and objective grading. The success of enterprise AI deployments hinges on factors like top-down buy-in, a dedicated 'tiger team,' well-defined evals, and the existence of 'scaffolding' or infrastructure for AI agents to interact with, contrasting with the relative maturity of physical autonomy systems like self-driving cars.