Quick Overview
Cooking with Codex is an OpenAI demonstration show hosted by developer content leads Dani Bauer and Alexander Behrens. Blending cooking instruction with developer tooling, the episode follows the hosts and invited creators as they test newly launched DevDay features. The presentation illustrates hands-on applications of Codex CLI, persistent agents, and voice models.
Key Points
- 1.OpenAI demonstrated persistent agents called dots that run continuously in the background to coordinate recipes, communications, and workflows.
- 2.The updated Codex command-line interface features a full-screen layout, agent management views, worktree branching, and duplex voice interaction.
- 3.A cooking sous-chef AI running on GPT Live 1 demonstrated full-duplex voice conversation, maintaining conversational state without interrupting mid-sentence pauses.
- 4.Guests used voice commands in Codex to generate an 8-bit ramen simulator game deployed directly to the ModRetro Chromatic handheld console.
- 5.A community showcase highlighted real-world AI applications, including a retro game built in thirty minutes and the decoding of a 217-year-old Napoleonic cipher.
Summary
Dani Bauer and Alexander Behrens introduce Cooked, an OpenAI video series blending culinary tasks with software development tools unveiled around DevDay. Preparing a ramen meal serves as the running theme for demonstrating agentic automation. Alex details how photographing kitchen ingredients allowed a persistent agent called a dot to assemble the recipe in the background. The hosts explain that dots operate continuously to manage notifications, check in guests, and draft correspondence. Alongside dots, ChatGPT Spaces serves as a shared canvas where agents and users write and organize plans, while Codex Cloud provides a dedicated cloud environment to run code independent of local machines.
A showcase of the refreshed Codex CLI introduces full-screen views, separate worktrees, an agent command center, and voice conversation capabilities inside the terminal. Back at the stove, an animated sous-chef avatar powered by GPT Live 1 assists with food preparation. Alex explains that GPT Live 1 is a full-duplex frontier voice model engineered to avoid interrupting the speaker when they pause or hesitate mid-sentence. The assistant gives precise instructions to simmer the eggs for 11 to 12 minutes and chop two cloves of garlic with ginger without browning them.
Matthew Berman, CEO of Forward Future, and editor Brian Knepprath join the kitchen. Matt shares how his dot proactively notified him the previous night about call sheets and filming instructions from his inbox. The group undertakes a challenge to build a playable game for the ModRetro Chromatic handheld console. Matt and Brian speak their requirements into the voice-enabled Codex CLI, requesting an 8-bit ramen cooking simulator with character selection based on dots. The model generates and runs the game in a terminal-based emulator, enabling them to navigate cooking stations and select ingredients.
OpenAI Communities Manager Federico Sarquis enters to host a community build review. The first featured project is Starfall, a retro side-scrolling shooter created by Cosmin Daiha in approximately 30 minutes using 6.1 So on Ultra mode. The second project is Carter Church's use of Astra to transcribe 1,300 hand-drawn signs and decode an unsolved Napoleonic military cipher that had resisted decryption for 217 years. The third project, by user bloomy btw, is a competitive horse racing game where the user-drawn sketch directly determines the running speed of the racer. The panel votes Carter Church's historical cipher decipherment the Catch of the Day.
To conclude the kitchen experiment, the hosts load their newly generated ramen simulator game directly onto physical ModRetro Chromatic hardware. Matt plays through the steps, moving a dot sprite across the kitchen, gathering ingredients, and boiling broth. Federico presents the completed bowl of cooked ramen, which the team samples together before signing off.
Cooking with Dots and Codex Cloud
Hosts Dani Bauer and Alexander Behrens kick off the episode by outlining features launched at OpenAI DevDay. They describe persistent background agents called dots, explain how ChatGPT Spaces provide a collaborative scratchpad for agents and users, and show how Codex Cloud enables developers to configure remote environments and work independently of local machines.
Codex CLI and GPT Live 1 Sous-Chef
An overview of the redesigned Codex CLI highlights a full-screen terminal interface, multi-agent tracking, separate worktrees, and native voice interaction. In the kitchen, a virtual sous-chef powered by the GPT Live 1 full-duplex voice model guides the hosts through cooking steps, pausing naturally and recognizing hesitant speech without cutting them off.
ModRetro Game Generation with Codex
Guests Matthew Berman and Brian Knepprath join the kitchen to participate in a one-shot challenge. Using voice input in the Codex terminal powered by an ultra-fast model, they prompt Codex to create an 8-bit ramen simulator game featuring character selection and cooking steps for the ModRetro Chromatic console, testing the result on an emulator.
Community Builds Showcase and Playtesting
Communities Manager Federico Sarquis presents three projects built with OpenAI tools, including the retro shooter Starfall, an Astra-assisted decipherment of a 217-year-old Napoleonic military message, and an interactive horse drawing game. The hosts crown the Napoleonic cipher as their favorite build before playtesting the newly coded ramen game on physical hardware.
The Bottom Line
The episode demonstrates how persistent AI agents, real-time voice interfaces, and automated code generation integrate into both casual interactions and software workflows. By successfully generating a playable retro game from conversational prompts and executing practical tasks, the showcase illustrates the practical capabilities of the Codex CLI and GPT Live 1 models. While presented in a lighthearted kitchen format, the video highlights expanding agent autonomy and multimodal computing in everyday development.
FAQ
What is Codex and how is Codex used in the Cooking with Codex video?
Codex is OpenAI's coding tool environment, demonstrated here through Codex Cloud and the redesigned Codex CLI, which enables developers to manage multi-agent worktrees and build software via terminal voice prompts.
What are persistent agents called dots according to the Cooking with Codex demonstration?
Dots are always-on, persistent AI agents that run in the background to handle tasks such as parsing inbox emails, checking in guests, and assembling recipes from ingredient photos.
How does the GPT Live 1 model handle voice interruptions and conversational pacing during live interactions?
GPT Live 1 is a full-duplex voice model that allows natural back-and-forth conversation, remaining silent when a user hesitates or trails off rather than mistakenly talking over them.
What features were introduced in the refreshed Codex command line interface demonstration?
The refreshed Codex CLI includes a full-screen layout, an agent command center for tracking background tasks, separate worktree branching, and built-in voice conversation directly in the terminal.
How did Carter Church use the Astra AI model in the community build featured on the show?
Carter Church used Astra to transcribe 1,300 hand-drawn signs and solve a 217-year-old encrypted military letter sent by Napoleon to his generals.
Worth watching for
Developers and AI practitioners interested in real-world agentic workflows, voice-driven coding, and OpenAI developer tools demonstrated through hands-on creative experiments.
- codex
- openai
- agents
- modretro
- chatgpt