OpenAI Just Merged ChatGPT and Codex. This Changes Everything.
Riley Brown · 2026-07-12
💡 Quick Take
1. OpenAI has integrated ChatGPT and Codex into a unified platform, introducing three new models led by GPT-5.6 Soul.
2. The newly released Soul model offers high token efficiency and strong tool-calling capabilities at a lower price point, though it is not positioned as a direct competitor to Anthropic's flagship Fable models.
3. Advanced agentic workflows like computer use allow models to execute tasks in the background, such as running end-to-end local software testing and UI evaluations.
4. Iterative feedback loops paired with well-defined success criteria enable agents to self-correct, grade their own outputs, and refine code or marketing scripts automatically.
5. New product features like native website hosting ("Sites") and a multi-tab in-app browser transform the platform into a comprehensive vibe-coding environment and agent-first operating system.
📊 Detailed Explanation
OpenAI's strategic shift merges the conversational interface of ChatGPT with the developer-focused capabilities of Codex, creating a super app designed to capture mainstream knowledge workers alongside power users. To support this broad audience, the interface includes targeted tabs, such as a "work tab" aimed at non-technical users looking to accomplish everyday professional tasks. This branding consolidation addresses awareness gaps by demonstrating that the ecosystem is fully equipped for complex, agentic workflows rather than simple text exchanges, making it a powerful vehicle for mass adoption.
Alongside the platform merger, OpenAI rolled out three new models: Soul, Terra, and Luna. Soul stands out as the primary engine for long-running, autonomous tasks and application development. Compared to its predecessor, GPT-5.5, the Soul model is significantly faster, more eager to complete entire plans without constantly asking for user confirmation, and highly token efficient. While top-tier frontier models like Fable remain unmatched in sheer raw intelligence for massive one-shot application building, Soul provides a remarkably cost-effective alternative for everyday development and complex multi-step execution.
A major operational leap discussed in the video involves computer use and automated feedback loops. Unlike older implementations that hijack the user's mouse and screen, the upgraded computer use feature in Codex can operate seamlessly in the background across multiple virtual instances and browser tabs. Developers and creators can harness this by setting up automated loops—such as instructing an agent to build a feature, test it via a browser interface, grade its own performance, and fix errors iteratively until a high success score is achieved. Marketers can apply identical logic by feeding high-performing reference materials, such as competitor ad transcripts or preferred design thumbnails, into the agent to guide generative tasks.
The platform also expands into hosting and modular tool integration. The newly accessible "Sites" feature allows users to prompt, build, and instantly host web applications directly within the app ecosystem. Furthermore, by sunsetting standalone browser projects and merging those engineering efforts into the in-app browser, OpenAI has solidified Codex as an agent-native operating system. Users can leverage external building blocks like Convex for databases, Daytona for sandboxes, and specialized MCP (Model Context Protocol) tools to assemble sophisticated software and marketing workflows entirely through conversational direction and parallel thread management.
🎯 Entertainment Expert Opinion
This discussion offers a compelling, high-level breakdown of how modern AI tooling is rapidly evolving from simple chat interfaces into autonomous software factories and agent-native operating systems. Riley Brown and Ross Mike successfully demystify complex technical concepts like multi-threaded execution, token efficiency, and computer use by grounding them in practical, real-world development and marketing scenarios. The conversation avoids overly dry technical jargon, maintaining an engaging, fast-paced dialogue that feels genuinely exploratory rather than purely promotional.
From a content quality perspective, the video is well-structured and highly informative, though it occasionally assumes the viewer is already familiar with specific industry shorthand and tooling ecosystems like Cursor, Replit, or various API frameworks. While the enthusiasm for OpenAI's latest releases is palpable, the hosts maintain a grounded perspective by acknowledging where competitor models—particularly Anthropic's Fable—still hold a decisive edge in raw intelligence. This balanced viewpoint enhances the credibility of their workflow recommendations, ensuring that viewers receive a realistic assessment of what current models can and cannot achieve.
The core takeaway for viewers is that the primary bottleneck in leveraging AI no longer lies in tool availability or model capability, but rather in human imagination and operational clarity. As the hosts emphasize, users must learn how to clearly define success criteria, establish feedback loops, and think like managers of software factories. Anyone looking to scale their productivity in coding, marketing, or general operations should watch this video to gain actionable insights into structuring agentic workflows, provided they are prepared to experiment beyond basic prompt engineering.
Kanal: Riley Brown