AI & I Podcast: "OpenAI's Codex: This Model Is So Fast It Changes How You Code"

Host: Dan Shipper
Guests: Thibaut (Head of Codex, OpenAI), Andrew (Technical Staff, Codex App, OpenAI)
Date: February 18, 2026

Episode Overview

This episode dives deep into OpenAI’s Codex app and its significance in revolutionizing the way developers—and even non-developers—interact with code. Dan Shipper interviews Thibaut and Andrew, two key leaders behind Codex, exploring strategic shifts, product decisions, user impact, and the breakthrough in speed that the latest Codex model (5.3) introduces. The conversation touches on the evolving landscape of programming tools, multi-agent capabilities, UX design, and the challenges and opportunities this new era brings.

Key Discussion Points & Insights

1. Strategic Shift: Expanding Codex’s Audience

• Super Bowl Ad & Mainstream Outreach

  • OpenAI aired a Super Bowl commercial for Codex (not ChatGPT), signaling a strategic move toward broader adoption.
  • Thibaut [02:36]: “We released the Codex app… more than a million downloads in the first week… It just felt that maybe we can inspire more people to build and show that agents are here. Right. It’s coming. It’s going to be mainstream.”
  • The ad served as a “love letter to builders” and created a surge of traffic and instant usage by viewers ([04:44]–[05:24]).

• Broader Target Group

  • Codex is no longer just for “professional engineers”—it’s now accessible to the technically curious, data scientists, and even less-technical users, thanks to usability improvements ([05:46], [07:45]).
  • Thibaut [07:45]: “Codex is like the most powerful experience right there… you should probably be able to read code in order to use Codex to its fullest.”

2. Product Design Philosophy: GUI vs. Terminal

• Why a GUI?

  • The team opted for a graphical user interface (GUI) over a traditional terminal (TUI), a notable departure from industry trends ([09:41]–[13:08]).
  • Thibaut [10:12]: “At some point it starts to feel very limiting, especially on multimodal… We felt like we needed to start experimenting with something else. And it is only when we saw it become super popular internally, where we're like, we have to ship this externally.”
  • GUIs allow for persistent/ephemeral panes, voice interaction, mermaid diagrams, full image rendering, and will facilitate more “dynamic stuff” in the future ([11:46], [13:08]).

• Breaking from Traditional IDEs

  • The Codex app offers a GUI for programming that is intentionally not an IDE.
  • Host [13:11]: “GUIs are great. IDEs are just the problem. There’s something that’s a GUI for programming that’s not an IDE, and it seems like you’re kind of in that, figuring that out, but I don’t even know what that’s called.”
  • Thibaut [13:46]: “It’s called a Codex app.”

• Adoption by Power Users

  • Even avid terminal users (including the guests and OpenAI team members) have found themselves preferring the Codex app GUI, especially for its ease in managing multiple agents and broader tasks ([15:51], [16:30]).
  • Andrew [15:51]: “Greg did an interview and Greg was like, ‘I am a power user. I thought I would never leave the terminal.’”

3. Breakthrough in Speed: Codex 5.3 Model

• Drastic Speed Improvements

  • The newly released Codex 5.3 is “so fast” that it fundamentally changes coding workflows ([18:06], [26:45], [29:09]).
  • Thibaut [18:10]: “I was surprised at how much faster it was… I have to adjust… I’m able to be more in the flow.”
  • Host [28:44]: “I think you guys recently put out a model that is so fucking fast… I can’t really keep up with this thing.”

• Speed as a Game Changer

  • The team had to intentionally slow down the UI so users could see responses populating ([29:09]–[29:33]).
  • Andrew [29:09]: “All of a sudden there was just, like, this wall of text… we need to smooth this thing out coming in. So we actually do slow it down ever so slightly.”
  • This acceleration unlocks new interactions—real-time agent collaboration, mid-turn steering, quicker code generation, and potentially voice-driven workflows ([30:07], [35:15]).

• Technological Infrastructure

  • Backed by partnership with Cerebras and significant backend infrastructure work (distributed systems, persistent websockets), these speed improvements will benefit current and future models ([31:52]).
  • Thibaut [31:52]: “We wrote the entire server stack to be based on websockets and a persistent connection… It decreases overall latency by like 30, 40%... We’re making the default for this new super fast model and then enabling on other models.”

4. Real-World Impact: Workflow Changes & Automations

• Evolving Workflow for Developers

  • Both guests now use the Codex app for nearly all development, leveraging its agent-oriented, multitasking capabilities ([07:01], [11:46]).
  • Andrew [07:01]: “99% of the code that I write is using the Codex app.”

• Automations & Novel Use Cases

  • Examples include:
    • Automated PR merging, conflict resolution, and issue fixing ([20:54], [21:49]).
    • Daily reports on code contributions ([22:00]).
    • Automated bug identification and fixing in random files ([22:37]).
    • Marketing research automations ([24:00]).
    • Creative applications: using the app to write a children’s book—including generating illustrations and assembling a PDF ([25:31]).
  • Thibaut [24:47]: “I love Andrew’s Yeet skill, which just takes the change and then does the commit, does the PR, writes the draft, puts it in draft and publishes a PR with a PR title and body.”

5. User Customization and Model Personality

• The Codex app allows users to toggle between “pragmatic” and “friendly” personalities, acknowledging that users want different styles from their coding agent ([27:08]).
• Thibaut [27:08]: “We obsess over [how the model feels]… Not everybody has the same preferences. The previous default was super blunt… Now we've introduced a more supportive, friendly personality, and you can just, like, pick between those.”

6. The Next Bottleneck: Verification and Code Review

• As code and feature generation accelerates, bottlenecks shift to:

  • Verification: Human validation of accuracy and compliance with requirements ([36:41]–[38:41]).
  • Code Review Overload: More generated code demands more review—posing challenges for teams ([37:51]).
  • Andrew [38:15]: “We’re taking a look at what that experience should look like with the model involved… review mode in the Codex app… annotates your diffs on the side with findings and stylistic things.”

• Speed Also Helps Understanding

  • Fast models assist in code comprehension and reduce review time ([38:41]).

7. Competitive Landscape & Internal Differentiation

• Learning from Anthropic & Claude Code
  • OpenAI observed and learned from Claude Code’s early lead, then focused on model reliability and rapid iteration ([41:22]).
  • In-house, the Codex team integrates product, engineering, and research, enabling them to ship improvements frequently and keep momentum high.
  • Thibaut [41:35]: “We had a lot of complaints on compaction… So we solved that end to end. And we decided to do end to end RL training and introduce compaction within research and then make the model itself very familiar with the concept…”

Memorable Quotes

• Thibaut [00:00]: “The first time I showed it to someone, they were like, no way. This is like a fake demo. This cannot be this fast. This will change everything.”
• Andrew [04:44]: “I’ve never seen a Linux CD in a Super Bowl ad. That was really cool to watch.”
• Host [13:11]: “GUIs are great. IDEs are just the problem.”
• Andrew [15:51]: “Greg did an interview… ‘I am a power user. I thought I would never leave the terminal.’”
• Thibaut [18:10]: “I was surprised at how much faster it was… I'm able to be more in the flow. That felt really good.”
• Andrew [25:31]: “I used the Codex app to make a book for my daughters… and then I printed it. So we’ve got like a super custom book.”
• Host [28:44]: “I think you guys recently put out a model that is so fucking fast. I can’t really keep up with this thing.”
• Thibaut [29:33]: “We actually do slow it down ever so slightly, just so that you can see the words come in, like, a little bit smoother.”
• Andrew [34:12]: “…if you have a model that's as almost as fast as running these scripts, then you can imagine a world where these things turn into skills… operations run a little differently with some intelligence…”

Notable Timestamps

• 01:26: Guest introductions
• 02:36: Codex app launch & Super Bowl ad—mainstream ambitions
• 07:01: Codex app as default workflow for OpenAI engineers
• 09:41: GUI vs. TUI: product strategy discussion
• 13:11 & 13:46: Host & Thibaut define the GUI/“non-IDE” approach
• 18:10: Impact of Codex 5.3’s speed on workflow
• 20:54: Automation use cases (PR merging, bug fixes)
• 25:31: Creative automation: generating a children’s book
• 27:08: Model “personality” options (pragmatic vs. friendly)
• 29:09: Model speed creates new UI/UX and workflow considerations
• 36:41: Speed shifts bottleneck from generation to verification
• 41:35: Reflections on Anthropic, compaction, and research collaboration

Episode Takeaways

• Codex is shifting Left: From tool for pro engineers to democratized platform for anyone technical.
• Speed is transformative: The new Codex model’s unprecedented speed enables new workflows, interactions, and possibilities—while forcing reconsideration of how verification, trust, and collaboration work.
• Automation is mainstream: Agents, automations, and skill chaining are not edge features but core to real-world productivity.
• UX experimentation is key: OpenAI isn’t dogmatic—terminal, GUI, and hybrid approaches are all being considered, but the GUI “app” stands out as the new baseline command center for agentic work.
• Verification is the new challenge: With code generation scaling, the friction shifts to validation, understanding, and code review.
• Culture of iteration: Tight collaboration between research, engineering, and product at OpenAI powers rapid progress.

For anyone involved in software development, AI tooling, or product strategy, this episode is a goldmine of practical insights and firsthand experiences from the cutting edge of AI-powered software creation.