AI agents are no longer that futuristic promise we used to hear about at conferences — they’re becoming a genuine part of everyday routines.
Codex, OpenAI’s programming and agent-powered work platform, is growing at a breakneck pace, and the numbers from a new report show this shift is already happening in a very concrete way.
The research was conducted in partnership with heavyweight American universities like Columbia, Duke, and the University of Pennsylvania, and it brought data that should catch the eye of anyone following the tech sector closely.
Artificial intelligence is moving out of conversation-and-web-search mode and into execution mode, and that changes a lot about how people are using these tools day to day. It’s no longer just about asking a question and getting an answer. Now it’s about handing off a task and letting the agent handle it while you do something else. 🤖
Why this matters
The leading AI labs spent years promising that efficient agents would work like little assistants taking care of our work, both at the office and at home. What used to sound like stage talk at tech events is now starting to become a tangible reality.
The report, shared first with Axios by OpenAI itself, shows that Codex usage is accelerating. And the most interesting part is the method the researchers used to reach that conclusion. They split Codex users into three broad categories: OpenAI employees, external organizations, and individual users. Then they measured how much each group uses Codex compared to ChatGPT, counting by output tokens.
What the report numbers revealed
The data is pretty eye-opening when you look closely. Check out how adoption varies from group to group:
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
- 99.8% of output tokens produced by OpenAI employees came from Codex.
- That number drops to 63% when we’re talking about external organizations.
- And it goes down to 16.5% among individual users.
OpenAI’s internal usage works almost like an ideal scenario. It represents how people could turn to agents when barriers like cost, access, training, and internal resistance essentially disappear. In other words, it’s a preview of how far things can go when friction is removed.
Among active ChatGPT and Codex users in organizations outside OpenAI, Codex adoption was sitting at just over 0% in August 2025. Today that share is already around 17%. That’s a massive jump in a short time, and it shows the adoption curve is climbing fast. 📈
Few users, but they use it a lot
One curious detail from the report is that the number of individual people using Codex is still small. But those who use it go all in. These users aren’t just testing the tool every now and then. They’re delegating real tasks — tasks that would take a human a significant amount of time.
In a sample of individual users, the numbers are impressive:
- 80.6% made at least one request to Codex estimated at more than 30 minutes of work for an experienced human.
- 70.2% made at least one request estimated to save more than an hour of human work.
- 25.6% went as far as delegating tasks that would take a human more than eight hours to complete.
Worth noting that these time thresholds are estimated by the model itself and based on a random sample of 0.1% of individual users who allowed their queries to be used for training. So yes, they’re estimates, but they give a solid sense of the scale of delegated work that’s already happening. ⏱️
Delegated work: the new way to use AI
The concept of delegated work is at the heart of this transformation. Instead of sitting in front of your computer guiding every step of a task, you describe what you need, set the parameters, and the agent executes while you turn your attention somewhere else. It sounds simple to explain, but in practice it represents a huge shift in the relationship between humans and machines. Artificial intelligence stops being a passive tool and becomes an active collaborator in the workflow.
Codex was built exactly for this. OpenAI positioned the platform not as an advanced chatbot, but as an agent infrastructure capable of operating autonomously in different contexts. And here’s a data point a lot of people didn’t see coming: non-developers are the fastest-growing user group, even though software work is still Codex’s core use case.
This usage model is already generating real impact. Professionals who don’t write a single line of code are discovering they can delegate administrative, organizational, and operational tasks to agents, freeing up valuable time in their daily routines. 🚀
When agents truly became part of the routine
The shift toward agentic work really kicked off in early 2026. That’s when everyday people started allowing tools like Codex, OpenClaw, and Anthropic’s Claude Code to interact with their computers on a deeper level.
In practice, this means letting agents:
- Manage calendars and appointments.
- Read and write files.
- Control web browsers.
- Run scripts and automations.
That’s a level of access that, not long ago, would have made a lot of people uneasy. And that initial skepticism makes sense, because handing control of your apps and files over to an AI isn’t a trivial decision.
From skepticism to daily use
There’s a pretty honest account of this transition from someone who has covered cybersecurity for years and whose coding knowledge stopped at early-2000s HTML. The natural reaction was caution when it came to giving agents access to files, browsers, and applications.
But over the course of a month, that same person started using Codex and Claude Code for a good chunk of the work and life admin they used to handle manually. The agents began filling out expense reports, organizing and triaging emails, booking hair salon appointments, and even reporting stolen packages at the building’s front desk. That last one, by the way, happens frequently enough in San Francisco to make automating it worthwhile. 😅
The psychological cost of getting started
One of the most interesting points raised about this movement comes from the world of workplace culture. According to expert Jessica Kriegel, agents are reducing what she calls the psychological cost of action.
In her view, agents make unfamiliar work feel more approachable. This means people start sooner, experiment more, and spend less energy worrying about what they don’t know. It’s a powerful observation, because it shows the benefit isn’t just about raw productivity. There’s also an emotional and confidence boost that helps people take the first step on tasks that previously seemed too complicated.
This factor might explain part of the growth of non-developers among users. When the psychological barrier drops, people who never imagined automating their own work start testing things out and getting surprised by the results. 💡
But hold on: most people are still just chatting
Despite all this progress, it’s important to keep things in perspective. The vast majority of AI users are still just having conversations with chatbots — not commanding an army of autonomous agents.
Among consumers on the Go, Free, Pro, and Plus plans who were active on ChatGPT or Codex in the past 28 days, less than 1% used Codex. In other words, heavy agent usage is still the domain of a select group, and there’s a massive adoption runway ahead.
What this means in practice
For anyone still watching from the sidelines, it’s worth understanding that adopting agents like Codex doesn’t require being an artificial intelligence expert. The whole point of the platform is to lower the barrier to entry for this kind of use. You don’t need to know how to code or understand the technical details of the model to start delegating tasks. What you need is the ability to clearly describe what you want and understand the limits of what the agent can handle on its own.
That practical understanding is what separates people who are getting real value from delegated work from those still using the tool superficially. The users showing up in the report data with the highest productivity are exactly the ones who found the right balance between autonomy and oversight — letting the agent run freely on the more operational parts and stepping in at the moments where human judgment makes the difference.
The trend is for this learning curve to get shorter and shorter as platforms evolve and users build up experience. Codex and OpenAI aren’t just building a technology — they’re building a new work habit that, by all indications, is here to stay. 🔧
