OpenAI’s Bold New ChatGPT Agent Can Now Do Complex Tasks

Key Points

• OpenAI launches ChatGPT agent for Pro, Plus, and Team users
• Agent performs tasks like planning, coding, research, and more
• New safety measures deployed to prevent misuse of capabilities
• Agent integrates with tools like Gmail, GitHub, and APIs

OpenAI has just launched its most advanced tool yet — the ChatGPT agent, designed to act more like a virtual assistant than just a chatbot. Rolled out on July 18, 2025, this general-purpose AI agent is now available to subscribers of OpenAI’s Pro, Plus, and Team plans.

Unlike earlier versions of ChatGPT that were mainly focused on answering questions, summarizing information, or generating content, the ChatGPT agent performs real tasks.

ChatGPT can now do work for you using its own computer.

Introducing ChatGPT agent—a unified agentic system combining Operator’s action-taking remote browser, deep research’s web synthesis, and ChatGPT’s conversational strengths. pic.twitter.com/7uN2Nc6nBQ

— OpenAI (@OpenAI) July 17, 2025

Whether it’s planning an event, creating a slide deck, writing and executing code, or even browsing the web for information, the agent is capable of doing all that and more.

Users can simply prompt the agent in natural language, such as:

• “Plan and buy ingredients for a Japanese breakfast for four.”

• “Research three competitors and build a slide deck for a meeting.”

• “Connect to my GitHub and review the latest pull requests.”

To activate it, users can now toggle “agent mode” from ChatGPT’s dropdown menu. This new feature combines OpenAI’s previous agentic tools like Operator, which could navigate websites, and Deep Research, which pulled structured insights from multiple sources, all into a single, cohesive agent.

🚨OpenAI just launched ChatGPT Agent (also called “Agent Mode”).

This new AI combines the capabilities of its earlier tools (Operator and Deep Research) into a unified agentic system that functions like a “virtual computer.”

Agent enables complex, multi-step workflows such as… pic.twitter.com/0dmC4ao9CI

— Brian Solis (@briansolis) July 17, 2025

This bold move follows a trend among AI developers, like Microsoft with its expanding Copilot Vision platform, and Google, which recently launched 8 AI-powered notebooks aimed at revolutionizing digital learning and productivity.

Deep tool integration makes the ChatGPT agent a true assistant

The ChatGPT agent doesn’t just respond with words — it takes action. OpenAI has enabled ChatGPT connectors, allowing it to plug into apps like Gmail, GitHub, and other APIs, giving it access to user data, files, emails, and codebases. That means it can now pull in data, filter content, or even act on emails and documents to help complete tasks on your behalf.

For developers, the most powerful addition is terminal access — the ability for the agent to write and run code in real-time. This opens the door for tasks like:

• Debugging software

• Running simulations

• Accessing APIs

• Automating internal workflows

OpenAI reports strong performance results for the model powering the agent. On the Humanity’s Last Exam, it scored 41.6%, nearly double the performance of OpenAI’s previous models (o3 and o4-mini). On FrontierMath, a tough benchmark for advanced reasoning, it scored 27.4% — far ahead of the former state-of-the-art.

These improvements suggest the agent isn’t just smart — it’s useful.

At the same time, this innovation comes just as OpenAI faces scrutiny over delays in releasing its open model roadmap, further fueling the conversation around transparency and competition in the AI space.

⁉️OpenAI Releases ChatGPT Agent

A unified AI system that combines deep research, operator capabilities, and more to autonomously complete complex, multi-step tasks using its own virtual computer.

The summarized points from the launch are in the comments 👇 pic.twitter.com/YB642RpGpk

— Nick Spisak (@NickSpisak_) July 17, 2025

Safety takes center stage with real-time monitoring

While the tech is exciting, OpenAI is treading carefully. The company says agent models pose unique risks, especially since they can interact with the real world.

In a detailed safety report, OpenAI labels the ChatGPT agent as “high capability” in sensitive domains such as biological and chemical threats. That doesn’t mean it poses a direct threat today, but the potential is there.

To address this, OpenAI has rolled out layered safeguards:

• Every prompt is scanned in real time

• If a request is biology-related, a second safety classifier is applied

• Responses that might evoke a biological threat are automatically blocked

Additionally, OpenAI has disabled the memory feature in the ChatGPT agent to reduce the risk of prompt injection attacks, where bad actors might manipulate memory to leak sensitive data. This is a notable change, as memory remains active in other ChatGPT experiences. OpenAI says it may reintroduce memory in the agent once it’s deemed safer.

OpenAI’s Preparedness Framework is also in full swing, marking this launch as part of its commitment to safe and responsible deployment of high-capability models.

The increased focus on AI safety reflects broader industry concerns. Other companies, like Grok AI, recently faced backlash for major blunders, including a code update linked to Nazi-related responses and public scandals requiring formal apologies.

How ChatGPT agent changes the AI product landscape

The launch of the ChatGPT agent marks a major moment in AI development. While many tech companies have pitched a future filled with helpful AI agents, the actual execution has mostly fallen short — early agents struggled with context, got stuck during execution, or offered unreliable outputs.

Now, OpenAI believes it has cracked the code.

By unifying capabilities like planning, reasoning, tool usage, and natural communication, the ChatGPT agent may finally deliver what others have only promised. It can complete multi-step tasks, adapt across domains, and learn from real-world inputs in a more fluid way — something very few AI agents have achieved at scale.

It also opens the door to deeper use cases:

• Business users can automate research and presentation creation

• Developers can code faster and debug more efficiently

• Every day users can plan trips, meals, or schedules with little effort

With agent mode, ChatGPT shifts from being a passive assistant to an active doer, hinting at a future where AI isn’t just responsive but truly interactive and autonomous.