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

Apoor Vagraval introduces Sherwin Wu and Olivier Godement from OpenAI, discussing the company's enterprise initiatives beyond ChatGPT. Sherwin highlights the challenges AI agents face due to the lack of existing infrastructure compared to physical autonomy, such as self-driving cars. Olivier emphasizes AI's potential in various industries, particularly healthcare, while Sherwin reflects on OpenAI's evolution from a B2B API aimed at building AGI to its current enterprise product line.

The conversation includes OpenAI's collaborations with traditional enterprises and digital natives, particularly following the rise of GPT models. An example with T-Mobile illustrates how OpenAI's models automate customer support, addressing high volumes of inquiries. The role of forward deployed engineers is emphasized, focusing on tailoring solutions and the importance of system design for effective model deployment. They discuss the challenges of connecting models to various tools and the need for clear success metrics.

The significance of Evals, especially audio evaluations, is highlighted in understanding customer experience quality. A new real-time API enhances the natural voice experience, with insights from the T-Mobile collaboration improving model performance. A partnership with Amgen aims to expedite drug development, focusing on managing vast data for R&D and administrative tasks, potentially benefiting millions. The collaboration with Los Alamos National Labs emphasizes a bespoke approach due to high-security requirements, targeting impactful scientific research.

The integration of OpenAI's reasoning model O3 into a secure environment on a supercomputer named Venado is discussed, showcasing the role of Forward Deployed Engineers. The high failure rate of AI deployments is noted, with successful initiatives requiring top-down buy-in and the establishment of a "tiger team" that combines technical skills with organizational knowledge. T-Mobile's approach of prioritizing AI initiatives while allowing teams to start small is presented as a successful model.

The conversation contrasts physical and digital autonomy, noting the structured environments of self-driving cars versus the unstructured settings of AI agents. Successful AI deployments require a platform to organize data, which many companies currently lack. The introduction of GPT-5 marks a significant advancement, focusing on improvements in instruction following and the ability to decline uncertain requests. Feedback on GPT-5 has been positive, particularly regarding its coding capabilities and reasoning, with a notable reduction in hallucinations.

Prompt engineering is emphasized as crucial for obtaining desired outcomes from GPT-5. The discussion also touches on the progress of companies benefiting from GPT-5 and the need for new evaluations for upgraded models. The announcement of a real-time API, featuring T-Mobile as a customer, marks a significant advancement, with progress in multimodal models noted.

The speaker expresses amazement at current models' capabilities, particularly in understanding accents and handling support calls. The discussion transitions to model customization, highlighting OpenAI's investment in reinforcement fine-tuning (RFT) for creating custom models. Base models are advancing in their ability to follow instructions, allowing for clearer descriptions of desired behaviors.

In business trends, eSports is identified as a significant growth area, particularly among younger audiences. The potential for AI to transform healthcare is highlighted, especially in streamlining processes within life sciences and pharmaceutical companies. The regulatory environment presents challenges, but also opportunities for AI to enhance efficiency.

Looking ahead, while the number of software engineers is expected to grow, the complexity of this increase is acknowledged. A poignant example is shared about a nonverbal individual who learned to create tools using ChatGPT, illustrating the transformative potential of these technologies. Sherwin emphasizes a global software shortage and advocates for customization in software development, predicting a shift in job roles where product managers will increasingly participate in coding.

For high school students, Sherwin advises prioritizing critical thinking skills, particularly in math and philosophy, as fields reliant on memorization may become less relevant. He encourages younger generations to embrace being AI-native, leveraging their familiarity with modern tools in the workplace. Reflecting on their journey at OpenAI, the speakers recount challenges that ultimately strengthened company culture and highlight the successful development and launch of GPT-5 as a major achievement.