How AI Agents Are Transforming Work: The OpenAI Report That Changes Everything

OpenAI has just published one of the most revealing studies on the real impact of artificial intelligence in the corporate environment. The document analyzes how AI agents — systems capable of autonomously executing tasks over long periods — are redefining the concept of knowledge work. And the numbers are, at the very least, impressive.

The End of the Era of Quick Response Chatbots

For years, interacting with AIs was dominated by short, point-in-time exchanges: you ask a question, and the AI responds, the cycle ends. This model, while useful, limited the transformative potential of the technology. Now, according to OpenAI, we are witnessing a pivotal shift.

Agentic AI changes the unit of knowledge work: from isolated interactions to long-horizon delegated tasks.

While a conventional chatbot operates in windows of seconds or a few minutes, an agent like Codex — OpenAI's programming and automation tool — can work independently for minutes or even hours, orchestrating tool calls, interacting with different digital environments, and continuously iterating in search of solutions. It’s the difference between asking for information and delegating an entire project.

The Real Experiment: OpenAI Itself as a Test Subject

What makes this report especially valuable is that OpenAI used its own company as a research lab. Internal data reveals a rapid and comprehensive transformation in the use of AI tools by employees.

By August 2025, the average OpenAI worker used Codex for less than 10% of their AI usage time. ChatGPT was still the standard tool for everyday work. But the scenario changed drastically in the following months.

Today, Codex accounts for an incredible 99.8% of weekly output tokens generated within OpenAI. This means that almost all AI-assisted content production in the company now goes through the agent, no longer through the traditional chatbot.

Four Trends Defining the New Paradigm

The study documents four major movements observed over the past year, both among individual and enterprise users as well as the OpenAI employees themselves:

1. Long-Horizon Tasks Have Become the Norm

Users are no longer asking for quick answers. They are delegating complex projects:

• 80.6% of sampled individual users made at least one request to Codex estimated to take more than 30 minutes of human work
• 70.2% made at least one request estimated to take more than 1 hour
• 25.6% made at least one request estimated to take more than 8 hours — equivalent to a full workday

Nearly a quarter of all requests to Codex involve tasks that would take more than one hour for a human to complete. Agents are literally working longer hours on more difficult tasks.

2. Codex Has Conquered All Departments of OpenAI

Adoption was not limited to engineering teams. According to the report, all departments at OpenAI are now using Codex as their primary AI tool for work.

• Engineering teams were the first to migrate
• The Legal, Finance, and Recruitment areas crossed the majority adoption point around April 2026
• Currently, Codex represents more than 85% of output tokens from the average OpenAI worker

This is a surprising statistic. Lawyers and recruiters using an agent AI tool as their main work resource would have been unthinkable two years ago.

3. The Explosion of Adoption Outside the Developer Universe

Perhaps the most revealing data from the study is the growth among non-developer users. Since August 2025:

• Among individual users: 137 times growth
• Among organizational users: 189 times growth
• Within OpenAI itself: 12 times growth

The pace of adoption by non-developers has far outstripped the growth among the programmers themselves. This indicates that the entry barrier to productively using AI agents has dropped drastically.

4. Professionals Performing Tasks Beyond Their Roles

One of the most fascinating effects documented in the study is the expansion of individual capabilities. Non-technical employees have begun using Codex to perform tasks that were previously outside their scope:

• Process automation
• Data transformation and analysis
• Creation and maintenance of internal tools
• Debugging code
• Structured analysis of complex information

In other words, the AI agent is functioning as a skill multiplier, allowing a Human Resources professional, for instance, to build an automation script without needing to hire a developer or learn programming from scratch.

What Does This Mean for the Future of Work?

OpenAI is clear in its interpretation: this pattern reflects what the company believes to be the future of work. The expansion of capabilities and accessibility of agentic tools will transform how professionals in all fields execute their functions.

The logic is simple but powerful: if an agent can work autonomously for hours, access tools, iterate on results, and deliver a final product — and any professional can delegate that work — then individual productivity is no longer limited by the number of hours in a day.

It’s a paradigm shift comparable to the arrival of the personal computer or the internet: it doesn't replace the professional, but exponentially amplifies what they can do.

A Transformation in Progress, Not a Future Promise

The most important point the OpenAI report makes clear is that this transformation is already happening. We are not talking about speculation about 2030 or 2035. The data is from 2025 and 2026, showing an accelerated and comprehensive adoption curve.

For companies, managers, and professionals across all fields, the message is direct: the era of AI agents has arrived, and those who learn to work with them — delegating complex tasks, validating results, and integrating these tools into the workflow — will have a significant competitive advantage.

The future of work is not about humans versus machines. It’s about humans who know how to use agents versus humans who do not.