Big Technology Podcast OpenAI President Greg Brockman on GPT-5.5 “Spud,” AI Model Moats, and Cybersecurity Risks
186 snips
Apr 23, 2026 
Greg Brockman, OpenAI president and co-founder, joins for a sharp look at GPT-5.5 “Spud.” They get into coding, spreadsheets, slides, browser use, and the push from chatbots toward real-world agents. Also on the table: model moats, distillation, cybersecurity risks, trust in agents, and why compute may become the economy’s key resource.
AI Snips
Chapters
Transcript
Episode notes
GPT 5.5 Crosses Into General Computer Work
- Greg Brockman says GPT-5.5 crosses from better coding into broadly useful computer work like slides, spreadsheets, browser actions, and end-to-end task completion.
- He frames it as a new way to use computers, with intelligence proactive enough to solve problems from minimal instruction.
OpenAI Shifted From Benchmarks To Agentic Work
- OpenAI now optimizes models for real business tasks, not just benchmarks, aiming for users to supervise fleets of agents rather than click every button themselves.
- Greg Brockman says finance, sales, and marketing workflows matter because leverage comes from abstracting away formulas, code, and UI steps.
GPT 5.5 Came From Whole Stack Co Design
- Greg Brockman rejects the idea that one training stage explains GPT-5.5, saying gains came from coordinated progress across pretraining, mid-training, reinforcement learning, data, and world connection.
- He compares it to building a car where a great engine alone fails if the rest of the system lags.