OpenAI Podcast: Coding Agents, GPT-5-Codex, and Future of Dev

Your friendly reminder that at bottom, the techniques for designing these technologies is autocomplete on steroids[bayesian probabilistic models, etc], and NOTHING close to replicating "agent" intelligence. Don't open your mouth for these hoodwinking stews. I suggest looking into actual biology, and maybe philosophy, to see how vague claims of agency are, especially visa vis claiming that software [which is what AI is] is "agent-complete". That's utterly metaphorical, but alas, someone needs to sell you stuff, so they'll make it seem like a piece of literal science

GPT-5-Codex has been released today. It's a special version of GPT-5 optimized for agentic coding. I'm pretty excited to give this a test run to see where we're at. The new model is intended to be able to do longer and more complicated agentic tasks like refactors or harder problems, i.e., real software engineering work. To try it, you'll need codex cli at least version v0.36.0, and you can now give it a spin with: `codex -m gpt-5-codex -c model_reasoning_effort="high"`

[Our Portfolio Company OpenAI Unveils New GPT-5 Model for Codex Programming Agents] Codex is a solution launched by OpenAI for AI coding agents such as GitHub Copilot and Claude Code, and it has received a major update yesterday.

🚀 *Big shift in software development ahead!* On Episode 6 of the *OpenAI Podcast*, Greg Brockman and Codex lead Thibault Sottiaux join Andrew Mayne to explore the leap from *code suggestions to coding agents*, the launch of *GPT-5 Codex*, and what the future of programming might look like by *2030*. 👉 Worth a listen if you're in tech, AI, or just curious about where we're headed. #AI #GPT5 #Codex #OpenAI #SoftwareDevelopment #FutureOfTech #CodingAgents

Moving from autocomplete to agent-complete marks a huge shift. Coding agents won’t just suggest snippets but handle workflows, reshaping how software gets built end-to-end. #Openai #AI

𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗾𝘂𝗲𝘀𝘁𝗶𝗼𝗻 𝗶𝘀𝗻’𝘁 𝘄𝗵𝗮𝘁 𝗔𝗜 𝗰𝗮𝗻 𝘀𝘂𝗴𝗴𝗲𝘀𝘁, 𝗜𝘁’𝘀 𝘄𝗵𝗮𝘁 𝗔𝗜 𝗰𝗮𝗻 𝗯𝘂𝗶𝗹𝗱 𝗮𝗹𝗼𝗻𝗴𝘀𝗶𝗱𝗲 𝘂𝘀. On Episode 6 of the OpenAI Podcast, Greg Brockman and Codex lead Thibault Sottiaux join Andrew Mayne to discuss the leap from code suggestions to coding agents, the launch of GPT-5 Codex, and what software development could look like by 2030. Watch now: https://openai.com/podcast

We’re excited to launch The Marco Show, a new JetBrains series exploring the big questions shaping software development. In the first episode, Marco speaks with Data Scientist Jodie Burchell about navigating the fast-changing landscape of large language models. From GPT and Claude to Gemini and beyond — what really matters when choosing the right one for your work? 📺 Watch the full episode on the IntelliJ IDEA YouTube channel and participate in a giveaway! https://lnkd.in/dwpj9FnR Also available on your favorite podcast platform.

“Schema gets stripped out.” True. But you need to ask: stripped out of what? Duane Forrester explained this on the latest episode of The Page 2 Podcast. Structured data is stripped out of the final LLM training tokens. The brackets, the code, the markup don’t carry forward. But before removal, the systems register what the schema means. The intent survives, even if the tags don’t. Think trimming fat: it’s cut away, but it still flavors the dish. That’s why dismissing schema because “AI ignores it” is a mistake. LLMs don’t need the brackets. They need the signposts of meaning. Your job isn’t to optimize for markup. It’s to optimize for semantic overlap (to be included) and semantic density (to pass the human sniff test). 👉 Full conversation with Duane starting around the 25-minute mark: https://lnkd.in/eb2cP3sA 💬 What are your thoughts on if / how LLMs use Schema?👇

If you are just about starting your career as a developer or even a writer and worried about how the AI tools like #Gemini may replace you, this episode from "Beyond the Code" short form podcast series featuring Christina Warren is for you ! Christina shares her PoV on how these tools can help you steepen your learning curve and enhance your productivity! Google for Developers Harsh Dattani Vania Noreen Scott Farish CV Paul Ravindranath G Sami Kizilbash

Check out Hugh Williams and Hannah Clayton-Langton's recently launched podcast, the cleverly named Tech Overflow Podcast. The podcast solves an incredibly important problem right now: as developing software and making technology becomes increasingly more accessible, how do non-engineers develop a scaffold so that they can begin to unpack what they know and don't know? The first episode discusses coding, explaining what code does at an abstract level and then systematically breaking down the mechanism. I enjoyed the discussion starting around 19:00, when Hugh and Hannah discussed the elements of what makes code beautiful and connected the concept to writing. If you're in tech and want to understand how software is made, from the technology, to the people and processes, I think Tech Overflow is going to be an amazing resource for you!