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

Feb 12, 2026 Episode Page ↗
Overview

Sherwin Wu, Head of Engineering for OpenAI's API and developer platform, discusses the evolving role of engineers and managers in the AI era, the future of AI models, and strategies for companies and startups to effectively leverage AI.

At a Glance
11 Insights
1h 19m Duration
16 Topics
5 Concepts

Deep Dive Analysis

AI's Impact on Engineering at OpenAI

The Evolving Role of Software Engineers with AI

The Sorcerer's Apprentice Analogy for AI Coding

Managing AI Agents and Contextual Challenges

Automating Code Reviews with AI

The Changing Role of Engineering Managers

Second and Third Order Effects of One-Person Billion-Dollar Startups

Management Lessons: Empowering Top Performers

Challenges and Best Practices in AI Deployments

Customer Feedback in AI: Building for Future Capabilities

Future Trajectory of AI Models (Next 12-18 Months)

Underrated Opportunity: Business Process Automation

OpenAI's Platform Strategy and Ecosystem Philosophy

OpenAI's Mission: Spreading AI Benefits to Humanity

Building on OpenAI's API and Developer Tools

Advice for Engaging with the AI Wave

Sorcery/Wizard Analogy for Programming

Programming is likened to sorcery where engineers are wizards casting incantations (programming languages) to make computers perform tasks. The current AI wave is the next evolution, with AI tools acting as literal incantations that execute desired actions, requiring skill to manage their powerful, high-leverage capabilities.

One-Person Billion-Dollar Startup (and effects)

The idea that a single individual can leverage AI to build a company worth a billion dollars. This implies a massive increase in individual agency and leverage, leading to a boom in smaller, specialized startups (e.g., 'one-person $10 million startups') and a potential shift in the venture capital ecosystem.

Models Will Eat Your Scaffolding for Breakfast

A principle stating that as AI models rapidly improve, the elaborate product scaffolding (like agent frameworks, vector stores) built to steer or support earlier, less capable models often becomes obsolete. Builders should anticipate future model capabilities rather than optimizing for current limitations, as the models will eventually handle tasks previously requiring complex external logic.

The Bitter Lesson (AI context)

An observation in AI and ML that less human-engineered complexity and more raw computation/data often lead to better scaling and growth. In building with AI, this means trusting the model's inherent capabilities more and reducing external logic or scaffolding, as the models tend to 'eat away' such architectural layers as they get smarter.

Overton Window (VC context)

The range of ideas tolerated in public discourse. In the AI space, the Overton window for venture capitalists has expanded, leading them to invest in competitive companies left and right due to the immense and rapidly growing market opportunity that is unlike anything seen before.

?
How much code at OpenAI is written and reviewed by AI?

The vast majority of code at OpenAI is authored by AI, with 95% of engineers using Codex daily and 100% of pull requests being reviewed by Codex, often reducing review times significantly.

?
How is the role of a software engineer changing with AI?

Engineers are becoming more like tech leads or managers, overseeing fleets of 10-20 parallel AI agents, steering them, and providing feedback, rather than primarily writing code themselves. This shift requires skill in managing and guiding AI tools.

?
What are common reasons for negative ROI in AI deployments?

Many companies experience negative ROI because AI deployments are often top-down mandates without sufficient bottoms-up adoption. This leads to a workforce that doesn't fully understand or effectively use the technology, as the implementation is divorced from actual daily work intricacies.

?
How should companies approach AI adoption for better ROI?

Companies should aim for both top-down buy-in and bottoms-up adoption. This can be achieved by forming an internal 'tiger team' of excited, technical-adjacent employees to explore AI capabilities, apply them to specific workflows, and facilitate knowledge sharing and excitement within the organization.

?
How should startups building on OpenAI's API think about competition from OpenAI?

Startups should focus on building valuable products that customers love, rather than worrying about OpenAI 'squashing' their ideas. The market opportunity is vast, and OpenAI views itself as a platform company committed to fostering an open ecosystem to spread AI's benefits to all of humanity.

?
What is OpenAI's mission regarding AI availability?

OpenAI's mission is to build AGI and spread its benefits to all of humanity. They achieve this by fostering an open ecosystem through their API, releasing models for developers, and providing free versions of tools like ChatGPT to democratize access globally.

?
What are the key trends for AI models in the next 12-18 months?

Expect models to coherently handle much longer tasks (multi-hour to day-long software engineering tasks) and significant improvements in multimodal audio capabilities, which is an underrated domain for business process automation.

1. Build for Future AI Models

When developing AI products, focus on where models are heading, not just where they are today, to create a product that will truly shine as capabilities improve.

2. Empower Top Performers

Managers should spend the majority of their time (over 50%) with their top 10% performers to unblock them and ensure they feel productive and heard, especially in the AI world where top performers can become exceptional.

3. Foster Bottom-Up AI Adoption

For successful AI deployment, combine top-down executive buy-in with bottoms-up employee adoption by staffing an internal ’tiger team’ to explore capabilities, share knowledge, and generate excitement.

4. Optimize AI Context for Agents

If your coding agents aren’t performing as desired, improve their context by adding more documentation, code comments, and structured information (e.g., .md files) to the codebase.

5. Automate Code Reviews with AI

Utilize AI tools like Codex to review 100% of pull requests, significantly reducing human review time and allowing engineers to focus on more complex tasks.

6. Anticipate Blockers with AI

Managers can use AI tools (e.g., ChatGPT with company knowledge) to proactively identify and unblock potential issues for their team, looking around corners to prevent future bottlenecks.

7. Leverage AI for Business Processes

Explore applying AI to automate and streamline repeatable business processes and operations, as this often overlooked area presents a massive opportunity for efficiency gains outside of traditional tech roles.

8. Focus on Customer Value, Not AI Labs

Startups should prioritize building products that customers genuinely love, rather than overly stressing about whether large AI labs will ‘squash’ their ideas, as the market opportunity is vast.

9. Engage with AI Tools Regularly

To avoid missing out on the current tech wave, actively lean into and use AI tools, understanding their capabilities and limitations, even if starting small.

10. Manage AI Information Overload

Avoid stress from the frenetic pace of AI news by focusing on engaging with just one or two tools initially, as you don’t need to know everything to participate effectively.

11. Adopt a ‘No Self-Pity’ Mindset

Cultivate a mindset of never feeling sorry for yourself, recognizing your agency to overcome challenges and pull yourself up in both work and life.

This is the worst the models will ever be.

Kevin Whale

The models will eat your scaffolding for breakfast.

Nicholas (Founder of FinTool, quoted by Sherwin Wu)

It literally feels like we're wizards now. You know, it feels like we're closer to to to having to making making it feel like this like magical experience where we're, you know, casting all these spells and having software do all these things for you.

Sherwin Wu

AI makes good people better and it makes great people exceptional.

Marc Andreessen (quoted by Lenny Rachitsky)

Never feel sorry for yourself.

Sherwin Wu

Effective AI Adoption Strategy for Companies

Sherwin Wu
  1. Ensure top-down buy-in from the C-suite, committing to becoming an AI-first company.
  2. Foster bottoms-up adoption by identifying or staffing a full-time 'tiger team' internally.
  3. This team should explore the full extent of AI capabilities relevant to the company's operations.
  4. Apply AI capabilities to specific workflows and business processes.
  5. Conduct knowledge sharing sessions and evangelize the technology to create excitement and facilitate learning within the organization.

Engineering Management Philosophy in the AI Era

Sherwin Wu
  1. Spend more than 50% of your time with your top 10% performers.
  2. Empower top performers by ensuring they have everything they need to do their work, acting as a supportive 'army'.
  3. Proactively look around corners to anticipate and unblock organizational or process-oriented bottlenecks for the team.
95%
Percentage of OpenAI engineers using Codex daily Indicates widespread adoption of AI coding tools internally.
100%
Percentage of OpenAI Pull Requests (PRs) reviewed by Codex Every code change merged into production is reviewed by AI.
70% more
Increase in PRs opened by engineers using Codex more The gap in productivity between AI power users and others is widening.
10 to 20 threads
Number of parallel AI agents managed by engineers at OpenAI Engineers are managing fleets of AI agents simultaneously.
From 10-15 minutes to 2-3 minutes
Reduction in code review time with Codex Codex significantly streamlines the code review process.
50% of the time
Model capability for multi-hour software engineering tasks Frontier models can coherently complete multi-hour tasks half the time.
80% of the time
Model capability for software engineering tasks under an hour Frontier models can coherently complete tasks under an hour most of the time.
800 million
ChatGPT weekly active users Reflects the massive global scale and adoption of ChatGPT.