OpenAI has upgraded its coding app, Codex, by introducing a dedicated macOS version with advanced multi-agent capabilities that enable users to manage multiple AI assistants working together to tackle complex tasks.
At its core, Codex started as a programming agent capable of writing code for users. However, the latest update allows it to delegate work among different models with complementary skills, effectively becoming a team of experts. This is evident in OpenAI's demonstration of creating a Mario Kart-like racing game from scratch, complete with visual assets and gameplay mechanics.
Codex leverages GPT Image for generating visuals while another model simultaneously codes the web game. The AI agent assumes multiple roles: designer, developer, and quality assurance tester, validating its work by playing the game itself.
To make it more accessible to users, OpenAI has introduced a "Skills" section that bundles instructions, resources, and scripts, allowing Codex to seamlessly connect with tools, run workflows, and complete tasks according to specific team preferences. Users can also choose to explicitly use specific skills or let Codex automatically detect them based on the task.
This new feature showcases Codex's potential in automating repetitive tasks, such as scheduling background work through its "Automations" section. OpenAI has been utilizing this functionality for tasks like issue triage and bug checking.
The release of Codex's macOS app aligns with the growing interest in exploring what AI agents can accomplish when working together in parallel. Recent examples include Anysphere's successful development of a web browser from scratch using this approach, although it encountered some issues along the way.
To give users a taste of Codex's capabilities, OpenAI is offering a limited-time opportunity to try the app for free, with ChatGPT Free and Go subscribers able to experience its potential. The company has also doubled rates for Plus and Pro subscribers in response to the new software's launch.
At its core, Codex started as a programming agent capable of writing code for users. However, the latest update allows it to delegate work among different models with complementary skills, effectively becoming a team of experts. This is evident in OpenAI's demonstration of creating a Mario Kart-like racing game from scratch, complete with visual assets and gameplay mechanics.
Codex leverages GPT Image for generating visuals while another model simultaneously codes the web game. The AI agent assumes multiple roles: designer, developer, and quality assurance tester, validating its work by playing the game itself.
To make it more accessible to users, OpenAI has introduced a "Skills" section that bundles instructions, resources, and scripts, allowing Codex to seamlessly connect with tools, run workflows, and complete tasks according to specific team preferences. Users can also choose to explicitly use specific skills or let Codex automatically detect them based on the task.
This new feature showcases Codex's potential in automating repetitive tasks, such as scheduling background work through its "Automations" section. OpenAI has been utilizing this functionality for tasks like issue triage and bug checking.
The release of Codex's macOS app aligns with the growing interest in exploring what AI agents can accomplish when working together in parallel. Recent examples include Anysphere's successful development of a web browser from scratch using this approach, although it encountered some issues along the way.
To give users a taste of Codex's capabilities, OpenAI is offering a limited-time opportunity to try the app for free, with ChatGPT Free and Go subscribers able to experience its potential. The company has also doubled rates for Plus and Pro subscribers in response to the new software's launch.