OpenAI readies ChatGPT superapp with agents and Codex

TechCrunch reported on June 7 at 9:23 a.m. PDT that OpenAI is still working on a major redesign of ChatGPT: a so-called super app that brings together chat, coding tools and AI agents. The report points back to the Financial Times, which said the changes are expected to start appearing in the coming weeks and are part of a wider push to make ChatGPT more useful for paid business work.

The shift matters because ChatGPT has mostly been understood as a place where people ask questions and receive answers. OpenAI now appears to be steering it toward a more active workspace. Instead of treating Codex, image generation, partner apps and agentic workflows as separate destinations, the company wants ChatGPT to become the front door for doing tasks across them. Thibault Sottiaux, who leads OpenAI's core product and platform, described the goal as a personal agent that can help across personal and professional life.

For everyday users, that could make ChatGPT feel less like a blank message box and more like an operating layer for work. A user might begin with a simple request, then be routed into code generation, research, design, scheduling or a connected third-party service. That is convenient, but it also raises the bar for trust. When an AI assistant moves from answering to acting, users need clearer controls over permissions, data access and mistakes.

For makers and companies, the move is a signal that the AI product battle is shifting from model quality alone to workflow ownership. OpenAI wants ChatGPT to be the place where people start tasks, pay for more capable tools and stay inside its ecosystem. Anthropic, Microsoft, Google and other AI providers are chasing similar territory with coding agents, workplace copilots and enterprise assistants.

The business reason is direct. Advanced AI is expensive to run, and OpenAI is under pressure to show stronger revenue before a possible listing. Turning free ChatGPT users into paying Codex, agent or enterprise customers could help. The open question is whether users want one AI hub for everything, or whether they will prefer specialized tools that are easier to understand, price and control.