Short answer: OpenAI's June 2026 data says knowledge workers are the fastest-growing group on Codex, adopting it about three times faster than engineers, yet only around 5 million people use it weekly against nearly 900 million on ChatGPT. So you are early. For a business owner, that gap is the opportunity, because this is bigger than a tool. Used properly it forces you to rebuild your operation around agents, the curve is steeper than ChatGPT ever was, and the owners who master it in the next 6 to 12 months will be operating from a different model before competitors notice. You have to go first, because you cannot delegate the thing that makes delegating safe.
There is a number from OpenAI this week worth sitting with. Codex, the part of the OpenAI stack that does work instead of just talking about it, crossed 5 million weekly active users. That sounds like a crowd until you put it next to ChatGPT, which is closing on 900 million. Roughly one in 180 of the people who open ChatGPT every week have crossed into the tool that actually executes. Everyone else is still typing into a chat box.
You have probably been told you are late to AI. On the thing that matters, owning agents that do real work, you are not late. You are early. And early is the rarest position a business owner ever gets to stand in.
The number that should reframe where you stand
ChatGPT reached about 900 million weekly active users and is on pace to pass a billion this year (TechCrunch). Codex passed 5 million weekly active users as of June 2, up more than six times since the desktop app launched in February (OpenAI). The tool that does the work is sitting at roughly half a percent of the tool that talks about it.
Read that gap as a map, not a scoreboard. Almost everyone who uses AI every day has not yet stepped from "I ask it questions" to "it does my tasks." That on-ramp is wide open and barely traveled, and the people already on it started this spring, not years ago.
"I'm not technical enough" is now factually wrong
The objection that keeps owners on the sidelines is some version of "that tool is for developers, not me." OpenAI's own data retired it this week. Knowledge workers, the analysts, marketers, operators, and researchers, now make up about 20 percent of Codex users and are adopting it roughly three times faster than engineers, with data-analysis use up 110 percent week over week (OpenAI). Knowledge workers is OpenAI's term. Business owner is my translation of it, after doing this work with more than ten of them. If the operators and analysts inside companies are pouring in this fast, that is your competition staffing up, and it is your opening.
Look at what shipped alongside that number. Six role-based business plugins, spanning sales, data analytics, creative production, product design, and finance. Sites, which lets someone stand up a working web workspace with no front-end developer. Annotations, which lets you point at a block of a spreadsheet and have the agent edit exactly that. None of it is built for engineers. When the company building the tool starts building for non-technical operators, "it's for developers" stops being a fact and becomes a story you tell yourself to stay put.
For an owner, this is not a new tool. It is a new operating model.
I have helped more than ten established small businesses implement AI agents and restructure their operations so agents can actually support how the business runs. So let me tell you the part the headlines skip. For an individual, Codex is a productivity boost. For a business owner it is a hundred times more consequential, because using it properly forces you to build an agentic business, and almost no one has done that yet. Do it and you are not in the early 5 million. You are in the first 1 percent.
The work is concrete, not theoretical. It looks like building the agent-ready business memory a whole team of agents can operate from, so they share context instead of guessing. It looks like client follow-up agents that work every lead without anyone having to remember. It looks like revenue and payment tracking agents that watch the money so the owner does not have to. None of that is a better prompt. All of it is operations, rebuilt so agents can run it.
Be honest about the climb, though, because this is where people underestimate it. This is not a ChatGPT learning curve, where a better prompt gets you a better answer. It is an operations curve. You are rewiring how work moves through your company: where your files live, what each agent is allowed to touch, how context is shared across a team of them, and how you know what they did. That is a much bigger lift, and the size of the curve is exactly why the lead it creates lasts. Businesses that master this in the next 6 to 12 months will not just be faster. They will be operating from a different business model before their competitors understand what changed.
Early is an advantage with a clock on it
Six times in four months is the steep part of a curve, not a plateau. And an operations advantage compounds, so the distance between the business that started and the business that waited does not grow in a straight line. It accelerates. Picture the owner across town who runs your kind of business and decides to start six months after you do.
- While they are still learning how to connect Codex to their apps, you already have agents running those apps on a schedule, without you in the loop.
- While they are still figuring out how to give their agents shared context as a team, you already have a team of agents working from one source of business memory, with clear rules for who touches what.
- While they are writing their first careful prompt, you are reviewing the work three agents finished overnight.
They are not less capable than you. They started two quarters later, on a curve where two quarters is the whole game. By the time they feel the pressure to catch up, you have compounded months of agents doing the work they are still doing by hand, and the competitor who used to be your equal is setting the pace in your market.
"I don't have time" and "can't my team handle this?"
Two objections come up every time, and both are traps.
The first is "I don't have time." You do not have time not to. The cost here is front-loaded and the payoff compounds, which is the best trade a business owner is ever offered. Waiting does not erase the cost. It moves it to later and raises the price, because then you are not building a lead, you are trying to close someone else's.
The second is "can't I just delegate this to my team?" No, and this is the one that catches sharp owners. You have to go first, because you cannot direct a transformation you have not felt. There are capabilities you cannot picture yet, and the only way you will see them is by doing the work yourself for a stretch.
Then there is the harder reason. Imagine everyone on your team has their own agent. Who knows which agent touched which file? Which agent made the mistake in the client report? Which agents can see your financials and which cannot? You will not know any of it, because a team of agents without infrastructure is not leverage, it is exposure. Someone has to build the foundation first: the access rules, the shared memory, the record of who did what. That someone is you, and you cannot delegate the building of the thing that makes delegating safe.
How to start today
Going first does not mean boiling the ocean. It means getting your own hands on it this week. Three setup moves, then one habit change that matters more than all of them.
- Download Codex to your desktop. Not a browser tab you mean to open later. The desktop app is where it can actually operate on your machine.
- Set up the sandbox correctly. This is the step almost everyone rushes, and it decides whether your agent does real work or stops to ask permission at every turn. Here is exactly how to set it.
- Connect your core tools. Your CRM like HubSpot, your meeting notetaker, the apps your business actually runs on. A plugin is the start, not the finish, so test each one before you lean on it. Begin with what a plugin really is, then install your first, and confirm your notetaker is actually agent-readable.
Then change one habit, and this is the part that does the real work. Force yourself to open Codex instead of ChatGPT. And force yourself to tell it to do the thing, not ask it about the thing.
You can feel the difference in one ordinary task. The chat habit is "can you write this email," and then you copy it, paste it, and send it yourself. The agent version is "write this email and send it, and when a reply comes in, draft my response and tell me." One hands you a draft and leaves the work on your desk. The other takes the task off your desk and comes back only when it needs you. That gap, between asking and delegating, is the entire shift this post is about.
If you do one thing this week, do this. Give Codex one real job today. Find one messy process in your business, point it at the files, have it produce the actual output, and tell it to leave a short record of what it changed. Do not ask it for advice on the process. Ask it to do the work.
Quick Answers
Is Codex only for developers? No. OpenAI's June 2026 data shows non-developers are about 20 percent of users and the fastest-growing group, adopting it roughly three times faster than engineers.
How many people use Codex versus ChatGPT? Codex passed 5 million weekly active users in June 2026, up more than 6x since February. ChatGPT has roughly 900 million. Codex is about half a percent of ChatGPT's base.
Can I just delegate this to my team? No. You go first to see what is possible, and you build the infrastructure, access rules, shared memory, a record of who did what, that makes a team of agents safe rather than chaotic. That part cannot be handed off.
How long until it pays off? It is an operations curve, not a prompting trick. Owners who master it in 6 to 12 months can be years ahead, because the advantage compounds and is hard to copy.
You are early. The only way to stay early is to go first.
Ready to give an agent its first real job? Start with the Codex sandbox setup, then install your first plugin.