“Engineers are becoming sorcerers” | The future of software development with OpenAI's Sherwin Wu

Key Concepts

• AI-First Transformation: OpenAI is undergoing a rapid shift to an “AI-first” company, requiring both executive support and grassroots adoption.
• AI as Augmentation, Not Replacement: AI tools like Codex are force multipliers for engineers, increasing productivity and changing the nature of their work.
• Rapid Model Evolution: AI models are evolving at an unprecedented pace, rendering existing tooling and strategies obsolete quickly. Building for future capabilities is crucial.
• B2B SaaS Opportunity: The ease of AI-powered startup creation is expected to fuel a boom in specialized B2B SaaS solutions.
• Context is Critical: Providing sufficient context and information to AI agents is essential for optimal performance.
• Multimodal Future: OpenAI is focused on expanding model capabilities, particularly in audio processing and handling longer, more complex tasks.

---

The Rise of AI in Software Engineering & Beyond

The tech landscape is experiencing a fundamental shift with the widespread adoption of AI, particularly OpenAI’s Codex. A striking statistic illustrates this: 95% of OpenAI engineers use Codex daily, and 100% of their pull requests (PRs) are reviewed by Codex. This isn’t simply about automating tasks; it’s changing the role of the software engineer, transforming them from primarily code writers to managers of “fleets of agents” – essentially, “wizards” issuing instructions and overseeing execution. This shift is increasing productivity, with engineers who use Codex opening 70% more PRs than those who don’t, and this gap is widening. Code review times have also been dramatically reduced, from 10-15 minutes to 2-3 minutes per PR with Codex assistance.

Navigating the AI-First Transition

OpenAI is actively transitioning to an “AI-first” company, emphasizing both top-down mandates and bottom-up engagement. Successful AI deployments require both executive buy-in and enthusiastic employee adoption. The recommendation is to establish internal “tiger teams” – dedicated groups responsible for exploring AI capabilities, applying them to specific workflows, and facilitating knowledge sharing. These teams don’t need to be exclusively comprised of software engineers; individuals with technical aptitude from various departments can quickly become proficient. A purely top-down approach, without grassroots adoption, is often ineffective.

The Speed of Innovation & Building for the Future

The AI landscape is evolving so rapidly that relying solely on customer feedback can be misleading. The phrase “the models will eat your scaffolding for breakfast” highlights how quickly advancements render previously essential tooling (like agent frameworks and vector stores) obsolete. In 2022, scaffolding was crucial for steering raw models; now, models are powerful enough that much of it is unnecessary. Therefore, companies should focus on building for the future capabilities of models, not current limitations. As Kevin Whale, VP of Science at OpenAI, famously stated, “This is the worst the models will ever be.”

Beyond Software: Business Process Automation & the Ecosystem

While Silicon Valley often focuses on open-ended knowledge work, a significant opportunity lies in business process automation (BPA). Applying AI to automate and improve repeatable, standardized processes offers substantial potential. OpenAI views itself as a platform company committed to fostering an ecosystem, releasing all models in its products through the API and not blocking competitors. The company advises startups to focus on building something people genuinely love, as the opportunity space is enormous. OpenAI’s mission is to spread the benefits of AI to all of humanity, necessitating a platform approach.

Current Focus & Future Directions

OpenAI is currently focused on increasing the length of tasks models can handle coherently (moving towards multi-hour tasks) and improving multimodal capabilities, particularly in audio processing. The “meter benchmark” tracks this progress; models currently handle multi-hour tasks 50% of the time and tasks under an hour 80% of the time. The API offers various levels of abstraction, from low-level access to models via the Responses API to higher-level tools like the Agents SDK and UI components. OpenAI currently has approximately 800 million weekly active users of ChatGPT.

---

Conclusion

The integration of AI, exemplified by tools like Codex, is fundamentally reshaping software engineering and creating new opportunities across various industries. Success in this evolving landscape requires a proactive, “AI-first” approach that combines executive vision with grassroots adoption, a focus on future capabilities rather than current limitations, and a commitment to building solutions that genuinely address user needs. The rapid pace of innovation demands continuous experimentation and adaptation, positioning those who embrace these changes to thrive in the emerging AI-powered world.