OpenAI Podcast Episode 6: The Future of Coding with AI

Date: September 15, 2025
Host: Andrew Mayne
Guests: Greg Brockman (OpenAI Co-founder and President), Thibaut Sochio (Codex Engineering Lead)

Episode Overview

In this episode, Andrew Mayne dives deep with Greg Brockman and Thibaut Sochio into how AI is rapidly transforming the art and practice of programming, the journey and evolution of Codex, the significance of building "harnesses" around models, the agentic future of coding (and society), and the just-released GPT-5 models for code. The group discusses both the technical evolution and the philosophical vision for the next decade, with practical examples, memorable quotes, and insights about what’s here, what’s coming, and why it matters for developers everywhere.

Key Discussion Points & Insights

1. The Origins and Evolution of AI Coding

• Early Attempts and Realizations
  • [00:59] Greg recalls GPT-3’s first code completions:

    "As soon as you saw that, you knew, this is going to work, this is going to be big. ...the goal of a thousand lines of coherent code has come and passed."

  • [01:45] Thibaut notes how developers quickly adapted to new possibilities:

    "It's incredible how used we get to things improving all the time... you reflect back to like a month ago, this wasn't even possible."

2. Specialization vs. Generalization in Models

• Greg discusses the tradeoff between building general versus specialized models—coding being a unique focus:
  • [02:04]

    "We're really here for the G, right? For AGI, general intelligence. ...Coding has always been the exception to that."

3. The Role of "Harnesses" in Usable AI

• What is a Harness?
  • [03:57] Thibaut explains:

    "You have the model... and the harness is how do we integrate that with the rest of the infrastructure so that the model can actually act on its environment. So it's the set of tools, it's the way that it's looping... the agent loop."

  • Analogizes:

    "Think about it a little bit as the harness being your body and the model being your brain."

4. From Prompt Completion to Agentic Coding

• The Shift in Interaction Paradigm
  • [07:41] Thibaut observes how users wanted more contextual, proactive help:

    "Instead of the user driving this thing, maybe let the model actually drive the interaction and find its own context... so that you can just sit back and watch the model do the work."

5. The Importance of Interface and Latency

• [05:11] Greg on product usability:

  "One thing that was very clear was latency was a product feature... anything that's slower than [1500ms], it could be incredibly brilliant, no one wants to sit around waiting for it."

• [15:00] Further explores the intelligence/convenience tradeoff:

  "There's intelligence... and then there's convenience... and there's some acceptance region where it's like, if the model's incredibly smart but it takes you like a month to run it... you still might..."

6. Agentic Coding and Its Modalities

• Discussing terminal, IDE, and async agents, and the 10X internal tool
  • [08:52-10:41] Thibaut:

    "...our goal is this entity, this collaborator that's working with you and then bringing that to you in the tools that you're already using as a developer."

  • Greg:

    "You need to kind of co-evolve the interfaces and the way that you use the model around its affordances."

7. Codex as a Collaborative Digital Entity

• [20:07] Greg:

  "Our vision is that there should be an AI that has access to its own computer... but is also able to look over your shoulder... These shouldn't be distinct things."

• [20:16] Thibaut:

  "This is also how we're thinking about Codex as an agentic entity that is really meant to just supercharge you when you're trying to achieve things."

8. AgentsMD & Codebase Navigation

• [20:39] Thibaut, on "AgentsMD" instructions:

  "It's like a compression thing... more efficient for the agent to just read Codex MD instead of exploring the entire code base... and then preferences that are not clear in the code base itself."

• [21:29] Greg:

  "How do you communicate to an agent that has no context what you want, what your preferences are?"

9. Competitive Landscape and Mission

• [22:43] Greg:

  "I focus a little less on the competition and a little more on the potential... we want AI to be available and accessible and benefit everyone."

• [24:12] Notes on future killer use-cases (e.g., automatic refactoring, tackling COBOL dependencies).

10. Real-World AI for Dev Workflows

• CLI, Code Review, and Developer Productivity
  • [25:22] Thibaut on reliable migrations:

    "Migrations are some of the worst things. Nobody wants to do migrations... If we can automate most of that, that's going to be a very beautiful contribution."

  • [26:48-28:20] Thibaut:

    "The big bottleneck for us was with increased amounts of code needing to be reviewed, the amount of reviews... We decided to really focus on a very high signal Codex mode... able to review a PR and really think deeply about the contract and the intention..."

  • [28:20] Greg on above-threshold AI capability:

    “Once you crack above some threshold of utility, suddenly people want it and get very upset if it gets taken away.”

11. The Arrival of GPT-5 Codex

• [30:36]
  • [30:47] Thibaut explains GPT-5 Codex:

    "GPT5 codecs is a version of GPT5 that we have optimized for codecs... it is one agent where you couple the model closely to the set of tools and it's able to be even more reliable... work, internally, up to seven hours for very complex refactorings."

  • [32:27] Greg on superhuman capabilities:

    "Even three, six months ago, I think our models were better than I am at navigating our internal code base to find a specific piece of functionality."

12. The Agentic Future: Millions of Agents, Human Oversight, and Safety

• [34:38] Thibaut on societal impact:

  "Large populations of agents... in the cloud... that we as humanity, as people, teams, organizations, supervise and steer... Incredibly important to solve is the safety, security, alignment..."

• [36:29] Greg on scalable trust:

  "How do you, as a human, manage agents that are out there writing lots of code? ...How do you maintain trust? How do you make sure that AI is producing things that are actually correct?"

13. The World in 2030: Where Are We Heading?

• [42:58] Greg’s bold prediction:

  "We will be in a world of material abundance... that'll be true in the physical world in addition to the digital world... But I think it'll be a world of absolute compute scarcity."

• [43:57] Thibaut on equitable access:

  "We're able to give it as part of the free, the plus plan, the pro plan, you can use codecs with your plus plan, you get GPT5, the same version that everyone else gets."

• [44:49] Greg on infrastructure challenges:

  "If we reach a point... where you're going to want agents running on your behalf constantly... now you're talking almost 10 billion GPUs that we need. We're orders of magnitude off of that."

14. Should You Still Learn to Code?

• [46:44]

  • Thibaut:

    "I think it's a wonderful time to learn to code."

  • Greg:

    "Yeah, I agree, definitely learn to code, but learn to use AI. That to me is the most important thing."

• [47:27] Greg on learning today versus a decade ago:

  "You're not going to have a tutorial that will flag this kind of issue for you. But will Codex in its code review be like, hey, there's JSON serialization. Just use this library? Absolutely."

15. Usage Trends and Closing Thoughts

• [49:17] Thibaut:

  "Usage has been exploding. So we've seen more than a 10x growth in usage from across users, and the users that were using it already are using it much more as well."

• [50:04] Greg on OpenAI’s mission:

  "I think we have so much to build. Progress continues on the exponential, and I think really bringing these tools to be usable and useful by everyone is core to our mission."

Memorable Quotes

• Greg Brockman:

  • "Bet that the greater intelligence will pan out in the long run." [00:19]
  • "Latency was a product feature." [05:11]
  • "Our vision is ... an AI that has access to its own computer ... but can also look over your shoulder." [20:07]
  • “Once you crack above some threshold of utility, suddenly people want it ... get very upset if it gets taken away.” [28:20]
  • "We will be in a world of material abundance... but a world of absolute compute scarcity." [42:58]

• Thibaut Sochio:

  • "The harness is your body, and the model is your brain." [03:57]
  • "It's the set of tools, ... the agent loop... you start to see pretty magical behavior." [03:57]
  • "GPT5 codecs... can work for seven hours on a hard refactor." [30:47]
  • "It's a wonderful time to learn to code." [46:44]

• Andrew Mayne:

  • "Reminded me of going into a grocery store and refusing to get a cart and just carrying everything to the checkout. ...Once you put things on wheels, it works really well." [17:02]

Major Timestamps and Segments

• Early Coding with AI & Model Progress: [00:59] – [02:04]
• What is a Harness? [03:54] – [04:38]
• Modality & Workflow Experimentation: [07:41] – [10:41]
• CLI vs. IDE, and the Evolution of Use: [17:49] – [20:07]
• Codex as an Agent/Entity: [20:16] – [22:25]
• Competitive Landscape & Mission: [22:43] – [25:11]
• Code Review & Dev Productivity: [26:48] – [30:29]
• Debuting GPT-5 for Coding: [30:29] – [34:19]
• Agentic Future Vision: [34:31] – [36:50]
• 2030 Outlook: [39:40] – [44:49]
• Learning to Code in the Age of AI: [46:35] – [48:41]
• Surging Usage: [49:09] – [50:00]

Final Thoughts

This episode offers a rare, candid, inside look at how OpenAI’s top minds see the present—and future—of programming with AI. The focus is not just on making individual developers faster, but on catalyzing entirely new kinds of software, teams, and economic structures built upon agentic collaboration, safety, and infrastructure. As model intelligence grows, the interface and its “harness” grows in equal importance, putting the human-AI partnership at the center of the 2030s software landscape.

For developers, teams, and entrepreneurs, the message is clear:
Now is a historic time to learn to code, but above all, to learn to wield and partner with AI. The future belongs to those who master harnesses, not just tools.