OpenAI: ChatGPT becomes an agent for serious work — operates for hours through apps and files until the job is done

On July 9, 2026, alongside the launch of the GPT-5.6 family, OpenAI announced ChatGPT Work — an extension that turns ChatGPT from a conversational assistant into an agent that independently completes work. An AI agent is a system that, given a goal, plans its own steps, executes them through tools and applications, and returns with a result, rather than waiting for instructions at each step.

What does OpenAI concretely promise?

According to the official announcement, ChatGPT can now “take actions across apps and files” and “work on a project for hours if needed,” turning a goal into completed work. The emphasis on multi-hour tasks is a key shift: previous agentic attempts (Operator, Deep Research) were limited to tens of minutes, while ChatGPT Work positions itself alongside tools like Claude Code that routinely handle long-running sessions.

Competitive context

OpenAI directly attacks with this product the segment where Anthropic (Claude agents in business tools), GitHub (Copilot coding agent), and Microsoft (Copilot agents in 365) are already active. The difference in approach: OpenAI builds the agent into ChatGPT itself — the product with the largest user base in the market — rather than into a separate developer tool. This brings agentic work to a mass, non-technical audience for the first time.

Open questions

The full article returns a 403 for automated access, so pricing, limitations, and the list of supported apps remain unknown until broader availability. It is also unclear which GPT-5.6 variant powers long-running sessions and how multi-hour tasks are billed. The same-day release alongside GPT-5.6 with multi-agent orchestration in the API suggests OpenAI is building its agentic strategy on two fronts simultaneously: for end users and for development teams.