The senior leaders responsible for guiding both the product vision and engineering strategy behind OpenAI’s developer platform have stressed that companies cannot successfully deploy artificial intelligence in the workplace without approaching the challenge with a strong focus on results and practical outcomes. According to Olivier Godement, who directs product development, and Sherwin Wu, who heads engineering, the process of embedding AI into day-to-day operations requires a deliberate strategy that balances executive commitment with employee empowerment. Their perspectives were shared during a conversation on the BG2 podcast released Thursday, where the two outlined a structured, three-part roadmap for moving from curiosity about AI to tangible business impact.
The first and perhaps most fundamental element Godement highlighted was the necessity of securing organizational buy-in from leadership while simultaneously creating a specialized group — often referred to as a “tiger team” — responsible for driving experimentation and implementation. This team, he explained, should not be narrowly confined to either external experts or in-house staff but should instead represent a blend of both. For instance, it might include technical experts from an AI provider like OpenAI working side by side with employees who bring deep insight into the company’s own business processes. In Godement’s view, this cross-functional mix is indispensable, as individuals with advanced technical knowledge can design and adapt AI solutions, while employees holding operational expertise are able to contextualize those solutions within the company’s unique workflows. He pointed out that in many areas — such as enterprise customer support — critical knowledge is often not formally documented but lives within the minds of employees. As a result, without a team combining technical capabilities and domain expertise, companies often struggle to translate AI’s theoretical potential into deployed solutions that deliver measurable value.
Building on this foundation, Godement and Wu explained that businesses must also establish explicit benchmarks, which they refer to as “evals,” to continuously measure whether AI is meeting operational expectations. These evaluations serve as a feedback mechanism to ensure momentum does not stall after initial enthusiasm. Godement cautioned, however, that constructing effective evals is considerably more complex than it may initially appear. Where leadership might expect straightforward dashboards, the reality is that meaningful measurements often need to emerge organically from employees themselves. Wu emphasized this point, noting that the most accurate evaluations are often sourced “bottom-up” from frontline staff, because they are the individuals most intimately aware of how AI influences daily tasks. Attempting to enforce such standards strictly from the top down, he warned, can miss the nuanced ways AI interacts with real-world workflows and therefore risks undermining successful adoption.
The final critical component described by the OpenAI executives is the need for organizations to monitor these benchmarks with rigor and persistence, while remaining flexible enough to adapt along the way. Godement noted that progress in adopting AI frequently feels less like an exact science and more like an evolving craft. In his view, managers and teams must cultivate a detailed understanding of how AI models are designed, how they behave under different circumstances, and where their limitations lie. At times, this may even require companies to fine-tune the models themselves, refining them to overcome clear shortcomings. Achieving success, according to Godement, is a matter of patience, iterative adjustment, and readiness to advance methodically until a solution is robust enough to integrate into business operations at scale.
Equally important, he added, is the organizational environment that leadership fosters. When senior executives make AI a priority topic and encourage employees to experiment, the entire company benefits. Creating space for teams to start small — perhaps piloting AI in limited use cases before scaling more broadly — allows employees to test, learn, and gradually expand their comfort level. This bottom-up experimentation, he explained, empowers employees to feel ownership of the transformation process, rather than experiencing it as a mandate imposed from above.
This philosophy aligns with an observable trend across technology companies, where CEOs are increasingly taking proactive steps to normalize AI experimentation within their organizations. Luis von Ahn, the CEO of language-learning platform Duolingo, told The New York Times in an August interview that his teams gather weekly for activities dedicated to exploring new AI applications. At Airtable, CEO Howie Liu has gone even further, publicly urging employees to treat hands-on engagement with AI as a high priority for personal and organizational growth. On a recent podcast, Liu even encouraged his teams to set aside normal responsibilities — including canceling meetings for days at a time — to experiment freely with AI tools that could enhance the platform’s long-term value.
Taken together, the advice of Godement and Wu illustrates that successfully weaving AI into the fabric of a business requires more than adopting new software. It represents an ongoing process of leadership commitment, structured evaluation, collaborative experimentation, and patient refinement. Companies that prioritize these elements stand a far greater chance of transforming artificial intelligence from an abstract concept into a powerful driver of operational efficiency, employee empowerment, and sustained innovation.
Sourse: https://www.businessinsider.com/openai-execs-3-things-companies-get-right-using-deploying-ai-2025-9
