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

Source: https://www.youtube.com/watch?v=yLTSqBzKG2s
Recap page: https://rapidrecap.app/video/yLTSqBzKG2s
Generated: 2025-09-11T22:02:41.716+00:00

---
## 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.

**Context:** This interview features Apoorv Agarwal speaking with Sherwin Wu, Head of Engineering, and Olivier Godement, Head of Product, for the OpenAI Platform. While OpenAI is widely known for ChatGPT, the discussion pivots to its enterprise-focused work, detailing how the company leverages its API and platform to serve businesses across various sectors like healthcare, telecommunications, and national security. The conversation explores the strategic importance of B2B offerings in fulfilling OpenAI's mission to distribute AGI benefits and dives into specific customer successes and future technological advancements.

## Detailed Analysis

OpenAI's enterprise division, spearheaded by Sherwin Wu and Olivier Godement, actively works to distribute the benefits of Artificial General Intelligence (AGI) through its platform, which includes a robust API, government solutions, and direct enterprise products. They emphasize that the API, OpenAI's original product, remains core to enabling developers and businesses. Key customer successes highlight the transformative impact of AI: T-Mobile utilizes OpenAI models for automating customer support, including voice interactions, with a focus on natural-sounding, low-latency responses, and has contributed to model improvements. Amgen, a healthcare leader, leverages GPT-5 to accelerate drug development and streamline regulatory processes, potentially impacting millions of lives. A notable deployment at Los Alamos National Labs involved a highly secure, on-premise solution on a supercomputer for national security research, demonstrating OpenAI's capability for bespoke, air-gapped environments. The discussion addresses the challenges and requirements for successful enterprise AI deployments, contrasting them with the more established 'scaffolding' in physical autonomy like self-driving cars. GPT-5 is presented as a significant advancement, offering superior intelligence, reasoning, and reduced hallucinations, though a trade-off between its deep reasoning capabilities and response latency is acknowledged. OpenAI has also innovated with a real-time API for voice, moving beyond traditional 'stitch' models to provide a more integrated and natural speech-to-speech experience. Furthermore, the platform offers advanced customization through Reinforcement Fine-Tuning (RFT), enabling customers to develop best-in-class models tailored to their specific needs by utilizing their own data and grading methodologies. The interview also touches upon the steep trajectory of AI agents compared to physical autonomy and the long-term vision for distributing AGI's benefits.

### OpenAI Enterprise Platform

- Focus on API and direct B2B solutions
- Mission to distribute AGI benefits
- Serving diverse industries like healthcare, telecom, national security

### Customer Success Stories

- T-Mobile for automated voice/text support, contributing to model improvements
- Amgen accelerating drug development and regulatory processes with GPT-5
- Los Alamos National Labs for secure, on-premise national security research

### GPT-5 Advancements

- "Amazingly intelligent" with improved reasoning and reduced hallucinations
- Trade-off between deep reasoning and latency
- "Top customer of GPT-5" used in healthcare acceleration

### Multimodality and Real-Time API

- Real-time API for natural voice interactions, surpassing "stitch model"
- Focus on improving voice model intelligence to match text capabilities
- Ongoing work to integrate orchestration logic across modalities

### Model Customization

- Investment in supervised fine-tuning and Reinforcement Fine-Tuning (RFT)
- RFT allows creation of "best-in-class" models using customer data and objective grading
- Examples: Rogo in finance, Accordance in tax

### AI vs. Physical Autonomy

- AI agents are "in day one here" with a "steep slope"
- Physical autonomy has more "scaffolding" (roads, stoplights)
- Success of AI deployments requires infrastructure and "scaffolding"

### Key Factors for Enterprise AI Success

- Top-down buy-in and "tiger team"
- Importance of "evals" (evaluations) for setting common goals
- Building "scaffolding" for AI agents to interact with