Podcast Summary: "Engineers are becoming sorcerers" | The future of software development with OpenAI’s Sherwin Wu

Podcast: Lenny's Podcast: Product | Career | Growth
Host: Lenny Rachitsky
Guest: Sherwin Wu, Head of Engineering for OpenAI’s API and Developer Platform
Date: February 12, 2026

Overview

In this episode, Lenny Rachitsky interviews Sherwin Wu, OpenAI’s Head of Engineering for the API and developer platform, about the rapidly shifting landscape of software engineering in the age of AI. Sherwin shares concrete insights from the inside of OpenAI, describing how tools like Codex have fundamentally changed the role of engineers, the shifting responsibilities of managers, and what’s coming next for startups, SaaS, and the future of productivity. The analogy of engineers as “sorcerers”—wielding AI agents like spells—frames a fascinating discussion full of actionable advice, tactics, and predictions about where AI development platforms and the broader economy are heading.

Key Discussion Points and Insights

1. AI as the New Standard in Code Production (03:21–07:27)

• Codex Usage: According to Sherwin, nearly all engineers at OpenAI use Codex daily, with almost 100% of code initially written by AI and 100% of PRs reviewed by Codex.
  • "95% of engineers use Codex. 100% of our PRs are reviewed by Codex." (00:00, 03:33)
• Productivity Leap: Engineers who use Codex heavily open 70% more PRs than those who do not, and this gap is widening.
• Changing Trust: Engineers’ trust in Codex grows day by day. The more they use it, the more responsibility they hand over to the models.

Quote:
"This is the worst the models will ever be, and so this is the worst that the models will ever be for software engineering as well."
— Sherwin Wu quoting Kevin Wheel (05:26, 01:02)

2. The Changing Role of Engineers: From Coders to Sorcerers (07:27–12:27)

• Engineers as Managers of AI Agents: Instead of hand-coding everything, engineers now coordinate dozens of AI-driven coding threads, giving feedback and direction—more like managing a team of apprentices.
  • “They're managing fleets and fleets of agents. It literally feels like we're wizards casting all these spells and these spells are kind of like going out and doing things for you.” (00:09, 07:27)
• SICP Metaphor: Sherwin references the classic programming book SICP ("Structure and Interpretation of Computer Programs") and its wizards/incantations metaphor, which finally feels real.
• Sorcerer's Apprentice Analogy: Like Mickey Mouse in Fantasia, engineers wield immense leverage—but must be careful not to let things go haywire without supervision.
  • "It's really powerful... you have to know what you're doing, right? In Sorcerer's Apprentice ... everything's flooding... it's at its greatest. And then the old sorcerer comes back and cleans everything up." (07:27)

Quote:
"It literally feels like we're wizards now... we're casting all these spells and having software do all these things for you."
— Sherwin Wu (07:27)

3. The Stress and Art of Orchestrating AI Agents (12:27–15:08)

• Agent Anxiety: There’s emerging stress when agents fail or get “stuck,” especially among engineers managing fully AI-generated codebases.
• Experiment Without an Escape Hatch: OpenAI has a team maintaining a 100% Codex-written codebase—no “manual override”—which surfaces issues like context specification, knowledge encoding, and limitations in agent autonomy.
  • "One interesting thing that we've noticed... a lot of the time when the coding agent is not doing what you want, it's usually a problem with context and just like information that you've given it." (12:43)

4. Automating the Boring: Code Review and CI/CD (15:08–18:51)

• Codex-Driven Reviews: Code reviews, once a tedious bottleneck, are now almost entirely automated. Codex reviews PRs, injects suggestions, and often allows engineers to skim more quickly or skip manual review on smaller changes.
  • "It makes code reviews go from a 10-15 minute task to sometimes even just a 2-3 minute task." (15:27)
• Automated Deployments: Linting errors, CI/CD steps, and other annoyances are handled by internal Codex-powered tools.

5. The Manager's New Role and Team Dynamics (19:28–24:14)

• Managers Are Next: While Codex isn’t yet a manager's tool, managers are using AI for performance reviews and admin tasks (e.g., aggregating work history from internal documentation).
• Leverage Top Performers: AI tools amplify productivity disparity—managers must focus even more on unblocking and empowering their top contributors.
  • “I've always leaned into top performers and spent more time with them, unblocked them, make sure they're happy.” (23:09)

Quote:
"AI makes good people better. Better and it makes great people exceptional."
— Mark Andreessen, quoted by Lenny (23:24)

6. The One-Person Billion Dollar Startup and Startup Boom (24:14–31:11)

• Second- and Third-Order Effects: The real revolution isn’t just “one-person unicorns,” but an explosion of micro-companies and SMBs making bespoke tools and B2B SaaS.
  • "There might be like a hundred other small startups building bespoke software that works extremely well to support other types of small, one person, billion dollar startups."
• Golden Age of Startups: If it’s easier to build, more people will do so—expect a golden age of SaaS and niche businesses.
• Support and Distribution Challenges: Solo founders may still face hurdles like customer support and distribution, potentially spawning further specialized products/companies.

7. AI Deployment in the Real World: Bottom-up vs. Top-down (37:59–43:56)

• Most AI Deployments Have Negative ROI: Outside of the Silicon Valley bubble, many organizations roll out AI top-down, without employee adoption or understanding, leading to poor results.
• Best Practice: Effective AI transformation needs both top-down support and bottom-up enthusiasm—ideally via an internal “tiger team” of skilled and motivated users to spread expertise.

8. Why "Listening to Customers" Can Be Misleading in AI (44:13–48:57)

• Field Changes Too Fast: Building what users want today may lead to dead ends as models accelerate and eat current “product scaffolding” (e.g., agent frameworks, vector stores, skills files).
  • "The models will eat your scaffolding for breakfast." (44:13)
• Key Lesson: Build for where the models are going, not where they are. Anticipate change, don’t just optimize for the present local maximum.

9. Predictions: Where Models and Platforms Are Going (50:16–53:17)

• Longer-Running Agents: Expect agents to coherently complete multi-hour (and soon, multi-day) tasks; this will enable very different product experiences.
  • "In the next 12, 18 months we could see models that could do multi hour long tasks very coherently." (50:34)
• Audio and Multimodality: Audio remains an untapped frontier; expect business process automation, customer service, and more to be impacted soon by huge leaps in audio models.
  • "A lot of the world's business is done via audio... I think that area is going to look very exciting in the next 12, 18 months." (50:34)

10. Business Process Automation: The Big Opportunity (53:47–56:36)

• Non-Engineering Work: Most of the economy runs on repeatable, deterministic business processes (SOPs), not open-ended engineering work—these are ripe for automation and AI transformation.
• Impact: AI could have even broader, less-discussed impact outside engineering—think support, operations, finance, etc.

11. OpenAI as an Ecosystem Platform (57:23–63:05)

• Startups vs. OpenAI: Sherwin urges founders not to fixate on being outcompeted by OpenAI; the opportunity is huge and the company maintains a platform-first, ecosystem-building approach.
  • "Just build something that people like and you will have a space in this." (57:49)
• OpenAI’s Charter: The company’s mission compels it to distribute AI’s benefits widely, empowering others to recreate solutions OpenAI couldn’t do alone. ChatGPT’s free tier, API releases, and upcoming app store underline this commitment.

12. Concrete Advice for Listeners (69:18–71:39)

• Don’t Miss the Wave: We are living through an extraordinary period of change; lean in, build, adopt tools, and don't take the opportunity for granted.
  • "I would just say engage with it... Just using the tools, understanding the limitations of what it can and cannot do so that you can kind of watch the trend of what it can start to do as the models improve." (69:32)
• Avoid Overwhelm: You don’t need to chase every new tool or trend—focus on learning and experimenting with a few, and don’t get caught up in the industry’s breakneck pace.

Notable Quotes

• "Engineers are becoming tech leads. They're managing fleets and fleets of agents. It literally feels like we're wizards casting all these spells and these spells are kind of like going out and doing things for you."
  — Sherwin Wu (00:09)

• "Make sure you're building for where the models are going and not where they are today... This is the worst the models will ever be."
  — Sherwin Wu (01:02, 49:07)

• "There might be a hundred other small startups building bespoke software that works extremely well to support other types of... one person, billion dollar startups. And so I think we might actually enter into a golden age of B2B SaaS."
  — Sherwin Wu (00:19, 24:30)

• "The models will eat your scaffolding for breakfast."
  — Sherwin Wu quoting Nicholas, Fintool (44:13)

• "My sense is I think managers will be able to manage much larger teams in this world... these tools will allow people managers to be higher leverage and will allow them to manage teams of way more than the current best practice."
  — Sherwin Wu (19:48)

• "The opportunity space in building with AI is so big... Just build something that people like and you will have a space in this."
  — Sherwin Wu (57:49)

Timestamps for Key Segments

| Timestamp | Segment / Topic | |---------------|-------------------------------------------------------------| | 00:00–05:18 | Codex adoption at OpenAI, code written/reviewed by AI | | 07:27–12:27 | Role shift: engineers as managers/sorcerers | | 12:27–15:08 | Agent orchestration stress, codebase experiments | | 15:08–18:51 | Automating code review, CI/CD, trust in Codex | | 19:28–24:14 | Management’s role, amplifying top performers | | 24:14–31:11 | The one-person billion-dollar startup & startup micro-boom | | 37:59–43:56 | Why many AI deployments fail, bottom-up change needed | | 44:13–49:07 | Customer feedback, “the models eat your scaffolding…” | | 50:16–53:17 | Where models & platforms are heading (multi-hour, audio) | | 53:47–56:36 | Business process automation as the next big frontier | | 57:23–63:05 | OpenAI’s ecosystem philosophy, platform focus | | 69:18–71:39 | Advice for listeners: lean in, don’t get overwhelmed |

Memorable Moments

• The Sorcerer's Apprentice Parallel (07:27): The metaphor becomes literal as engineers now command fleets of semi-autonomous agents, needing vigilance to avoid chaos.
• “This is the worst the models will ever be” (05:26, 01:02): A mindset shift that compels everyone to look ahead, not optimize for the present.
• Golden Age of Bespoke SaaS (24:30): Forecasting a Cambrian explosion of software built by micro-firms to enable other micro-firms.
• "The models will eat your scaffolding for breakfast." (44:13): A warning against over-engineering for today’s limitations.
• “Never feel sorry for yourself” (75:59): Sherwin’s personal motto for work and life.

In Sherwin’s Own Words (Selected Quotes by Timestamp)

• On AI’s capabilities and trust:
  "Every day, I talk to someone who is blown away by something it can do and their bar of trust... goes up over time." (05:26)

• On leveraging the current era:
  "I think the next two to three years are going to be some of the most fun in tech and in the startup world that we’ll have in a very long time. I would just encourage people not take it for granted." (68:26)

• Practical advice:
  "Just using the tools, understanding the limitations... just leaning into like one or two different tools, starting small, is already more than you need here." (71:39)

Conclusion

This episode paints a vivid portrait of the present and near future of software engineering, management, and entrepreneurship in the AI era. From engineers as “sorcerers” to the coming explosion of tailor-made SaaS and the importance of not standing still or freezing from overwhelm, Sherwin Wu’s worldview is pragmatic, optimistic, and empowering. The message is clear: the AI transformation is here; engage, experiment, and help shape what comes next.

For more:
Find Sherwin Wu on X/Twitter @erwinwu
Explore more episodes and resources at lennyspodcast.com