ChatGPT Work: OpenAI turns ChatGPT from conversational assistant into a real operational agent with GPT-5.6 and Codex

There is a question that deserves to be the starting point of this analysis: a few years from now, will we still use AI to ask questions, or to delegate entire days of work to it? The answer OpenAI is giving with ChatGPT Work is clear: the latter. And they are doing it not by building a new stand-alone product from scratch, but by integrating advanced agentic capabilities directly inside ChatGPT — the most used AI platform in the world — making them accessible to a base of hundreds of millions of users who have never heard of "AI agents" or automation frameworks. This is OpenAI's most important strategic move of the last few months. Not a more powerful model. Not a new interface. But the transformation of the world's most popular consumer product from a question-and-answer tool into an autonomous operational agent.

ChatGPT Work is ChatGPT's new operational mode, powered by GPT-5.6 and Codex, that transforms the fundamental way you interact with the system. Instead of asking one question at a time and receiving an answer to read, you assign a goal and ChatGPT Work works autonomously to achieve it — across documents, presentations, spreadsheets, web searches and activities distributed across multiple applications — without requiring continuous user supervision. The conceptual distinction is fundamental. Classic ChatGPT: "write a risk analysis for this product", then read the answer, copy, paste, adjust, open Excel, enter the data, open PowerPoint, create the slide. ChatGPT Work: "starting from this product sketch, find the main suppliers on the market, analyze the risks, build an Excel workbook with the data and generate a complete presentation using our corporate template", then go do something else and come back when it is done. This is not a difference in features but a difference in the relationship between user and AI: in the first case AI is a tool that amplifies your abilities but requires your continuous presence, in the second it is a delegate that autonomously executes complex work while you do something else.

ChatGPT Work is not built on a single model: it is powered by the combination of GPT-5.6 and Codex, two systems with distinct and complementary roles. GPT-5.6 is the new version of OpenAI's flagship model, specifically optimized for agentic tasks — planning sequences of actions, reasoning about long-term goals, managing complex contexts that span multiple documents and applications, and deciding autonomously how to proceed when facing ambiguities or obstacles. Compared to previous versions, GPT-5.6 shows significant improvements in multi-step planning with dependencies: the ability to decompose a complex goal into a sequence of steps with correct dependencies ("first find the suppliers, then analyze prices, then build the comparison, then create the presentation") and to execute this sequence coherently, adapting when the results of one step affect the ones that follow.

An equally important improvement is in error handling and autonomous recovery. When an action fails — a document does not open, a search does not produce useful results, an application responds in an unexpected way — GPT-5.6 does not stop and does not ask for help: it analyzes the error, identifies the cause and finds an alternative path to reach the goal. It is this autonomous recovery capability that distinguishes a real agent from a simple automated script. A third key element: understanding intent beyond literal instructions. An agent that executes instructions in a pedantically literal way is as dangerous as one that ignores them. GPT-5.6 has significantly improved its ability to understand the user's real intent — the "why" behind the instructions — and to make autonomous decisions that respect that intent even when the literal instructions are ambiguous or incomplete.

If GPT-5.6 is the planning brain, Codex is the executive arm: the system that translates plans into concrete actions on tools and applications. Codex is OpenAI's model specialized in interaction with external software and systems — capable of writing and executing code, calling APIs, navigating web interfaces, manipulating files and documents, and communicating with external applications through their native tools. In the ChatGPT Work pipeline, Codex receives GPT-5.6's instructions ("open the corporate PowerPoint template and insert the workbook data into slides 3, 5 and 7") and translates them into concrete actions: writing the necessary code, calling the appropriate APIs, or directly manipulating files in the required format. The GPT-5.6 + Codex combination creates a system where high-level reasoning and low-level execution are perfectly integrated, overcoming the main limit of previous agentic systems where "thinking" and "acting" were often misaligned.

In the official launch video, OpenAI shows a scenario that perfectly summarizes the system's capabilities. The initial input is a hand-drawn sketch of a product idea, photographed and uploaded into ChatGPT Work, with a goal in natural language: "starting from this concept, prepare everything I need to evaluate whether it is worth developing". What ChatGPT Work does autonomously: it analyzes the sketch, understanding the product concept, identifying the main features and classifying the market category; it runs structured web searches to identify the main suppliers of the necessary components or services, gathering information on prices, availability and reliability; it builds a structured analysis of the risks associated with developing the product — market, technical, regulatory, financial — based on the information gathered; it creates a structured Excel sheet with supplier comparison, risk matrix and preliminary cost projections; finally it opens the corporate PowerPoint template, populates the slides with the processed information, adapts the layout to the specific content and produces a finished, professional presentation. Total time: a few minutes, with no user intervention between the initial input and the final result. This is not an artificially simplified lab scenario: it is exactly the kind of task a junior analyst could spend half a day completing, and that ChatGPT Work executes autonomously in minutes.

Anyone already familiar with Anthropic's Claude Cowork — the product that introduced similar concepts to the market — will find ChatGPT Work familiar in its basic approach, but there are significant differences that reflect different philosophies. Claude Cowork bet on depth of integration with a controlled ecosystem: specific tools, defined workflows, high-quality reasoning on complex tasks but in circumscribed environments. Anthropic's approach was more cautious, more control-oriented, focused on the quality of the single agentic task. ChatGPT Work bets on breadth: integrations with a much larger number of applications and tools, accessibility to a much wider audience, a user experience designed for people who are not necessarily technical power users. OpenAI's approach is more aggressive on distribution, more oriented toward mass adoption. The most practically relevant difference is that ChatGPT Work is integrated directly into ChatGPT — the product that hundreds of millions of people already use daily. It does not require learning a new tool, changing workflow or signing up for a new service. Claude Cowork, however technically sophisticated, requires a tool switch. ChatGPT Work does not. And in the battle for mass adoption, zero friction matters enormously.

OpenAI has announced native integrations with a broad ecosystem. The Microsoft 365 suite is the deepest and most strategically significant integration — unsurprising, given the historical partnership with Microsoft: ChatGPT Work can natively read, create and modify Word, Excel, PowerPoint, Outlook and Teams, without conversions or workarounds. Google Workspace is supported at a comparable level — Docs, Sheets, Slides, Gmail and Drive — allowing users who live in the Google ecosystem to benefit from the same agentic capabilities. Web search is deeply integrated: ChatGPT Work can run structured searches, analyze pages, extract data from websites and synthesize information from multiple sources, weaving the results directly into the current task workflow. File management covers PDFs, images, CSVs, JSON and source code, with the ability to navigate complex folder structures. For enterprise users, integration with custom APIs and internal corporate tools is supported. Finally, native integrations with project management tools like Notion, Asana and Linear allow ChatGPT Work to create tasks, update statuses and synchronize information between different systems.

At launch ChatGPT Work is available for Pro users with full access to all features, for Enterprise users with additional governance and data control features, and for Education users with full access for academic institutions. Rollout for Plus and Business users is planned within the next few days — a very tight time window that signals the priority OpenAI attributes to the rapid spread of this feature. Free users do not yet have a confirmed date for access, but OpenAI's history suggests that features are progressively rolled down to lower tiers in the months following launch. The speed of the rollout is itself a strategic signal: OpenAI is not treating ChatGPT Work as a premium experiment, but as a core feature meant to define the standard ChatGPT experience in the coming years.

ChatGPT Work is not simply a productivity tool: it is the first mainstream signal of a structural change in the way knowledge-intensive work is organized and executed. A huge class of professional work consists of gathering information from different sources, synthesizing it into a structured format and presenting it accessibly. Reports, analyses, market research, due diligence, competitive benchmarks: all of this is exactly the kind of work ChatGPT Work can execute autonomously with increasing quality. People who do this kind of work exclusively will have to redefine their added value. Preparing meetings, synthesizing documents in advance, producing briefing materials, updating presentations with fresh data are another category of high-volume low-added-value work that ChatGPT Work fully absorbs. Structured administrative work — filling forms, updating databases, generating recurring reports, syncing information between systems — is a direct candidate for automation.

What is left for humans? The question every professional should be asking is not "will ChatGPT Work do my job?" but "which parts of my job stay mine when the agent autonomously executes the rest?". The emerging answer is clear: defining goals — deciding what is worth doing and why — stays human; judgment in ambiguous situations where data alone is insufficient stays human; the human relationship — negotiation, persuasion, care, leadership — stays human; strategic creativity, the ability to see non-obvious connections and reframe problems in ways that unlock new solutions, stays human. The historical pattern is clear: every time a technology automates a level of work, value shifts to the level immediately above. ChatGPT Work automates the executive layer of knowledge work; value shifts toward goal definition, critical judgment of the results produced by the agent, and the ability to orchestrate multiple agents on complex projects.

For those building AI products, the strategic lesson of ChatGPT Work is even sharper. The window to build wrapper products around a single model — a vertical chatbot, an assistant for a specific sector — is closing rapidly. What becomes defensible is the orchestration layer, deep integration with the company's proprietary data and governance over agents that execute actions with real effects on corporate systems. For users the lesson is simpler: over the coming months it pays to invest time in learning how to formulate well-written goals — clear, measurable, context-rich — because this skill will rapidly replace the ability to write good prompts as the most valuable AI skill on the job market. ChatGPT Work is not the end of human work: it is the end of a specific mode of work. Those who understand the difference in time will find themselves working with a tireless delegate that executes in minutes what previously required days. Those who do not will find themselves competing with that delegate to do the same things.