“Engineers are becoming sorcerers” | The future of software development with OpenAI’s Sherwin Wu
“Engineers are becoming sorcerers” | The future of software development with OpenAI’s Sherwin Wu
“Engineers are becoming sorcerers” | The future of software development with OpenAI’s Sherwin Wu
Sherwin Wu, who leads engineering for OpenAI’s API platform, discusses how AI is rapidly transforming software development, team dynamics, and startup economics—not as a distant future, but as an operational reality happening now inside OpenAI and across the industry.
At OpenAI, Codex generates nearly all code and reviews every PR, cutting review times from 10–15 minutes to just 2–3 minutes. Engineers are shifting from writing code line-by-line to orchestrating autonomous AI agents—requiring precision in prompting and oversight, much like a sorcerer managing powerful, literal-minded genies. This transition is widening the productivity gap between AI power users and others, while also redefining engineering management: top performers now receive disproportionate support, treated like surgeons needing frictionless focus. AI enables 'one-person billion-dollar startups,' but success increasingly hinges on distribution and audience ownership—not just technical execution. Critically, developers must build for the trajectory of models, not today’s scaffolding; as Nicolas Bustamante warns, 'models will eat your scaffolding for breakfast.' Over the next 12–24 months, expect major advances in long-context reasoning, multimodal (especially audio) interfaces, and business process automation beyond engineering. OpenAI’s platform strategy remains deliberately open—prioritizing broad API access, ecosystem growth, and democratized AI access—to accelerate global impact.
00:00
00:00
95% of engineers use Codex for coding and 100% of PRs are reviewed by it
03:15
03:15
Nearly 100% of code is generated by AI, with 95% of engineers using Codex daily and 100% of PRs reviewed by it
09:26
09:26
AI tools act like genies—granting wishes literally, so instructions must be precise to avoid unintended consequences
12:27
12:27
Removing the option to not use AI has helped identify problems that need solving for greater agent adoption
17:29
17:29
Almost every engineer at OpenAI heavily uses Codex, and the vast majority of code is likely authored by AI
22:16
22:16
AI makes good people better and great people exceptional
29:28
29:28
Distribution becomes more important as there are many things vying for attention, making those with an audience and platform more valuable
31:45
31:45
Spend over 50% of management time with top 10% performers to unblock their progress
40:11
40:11
Tools like Codex at OpenAI drove internal adoption by getting employees excited
43:57
43:57
Models will eat your scaffolding for breakfast
48:57
48:57
Build for where models are going, not where they are today
52:56
52:56
Audio/speech will become more important and integral to the AI experience
53:35
53:35
There's a large, underrated opportunity to use AI to automate repeatable business processes, integrate with business data and systems, and improve productivity and work methods
57:23
57:23
OpenAI views itself as an ecosystem platform company, releases all models through its API, aims to keep the platform neutral, and supports competition
1:03:11
1:03:11
For $20 a month, users get the same Pro model as billionaires, enabling unprecedented AI democratization
1:05:22
1:05:22
The Responses API is popular for building long-running agents
1:15:39
1:15:39
A life motto of not feeling sorry for oneself
