The Spreadsheet Precedent: Why Codex Is the VisiCalc Moment for Coding
The spreadsheet turned every office worker into an accountant. Codex is about to do the same to coding — and the people most at risk are not the engineers.
OpenAI published a report this morning that reads like a product update. Buried inside it is a number that should make anyone in an office sit up: knowledge workers now represent roughly one-fifth of Codex's users, and that segment is growing three times faster than the developer population. In March, Fortune reported that Codex had 1.6 million weekly active users. As of this week, it has crossed five million. The desktop app that drove that growth launched in February.
Axios first reported the details of OpenAI's own data. The fastest-growing knowledge-worker task is data analysis, up 110 percent week over week. Research is up 37 percent. Knowledge artifact creation — reports, memos, contracts, spreadsheets — is up 36 percent. These are not coding tasks. These are the tasks that define a professional's judgment: synthesizing information, building arguments, producing deliverables that took years to learn how to produce well.
The Axios story called this a productivity story. It is not. It is a deskilling story — except the people being deskilled do not have a name for it yet.
OpenAI would frame this as empowerment. The data supports a more uncomfortable reading. The tasks growing fastest are precisely the ones that used to require a human in the loop to catch the errors the model makes.
Andrew Hall, a professor at Stanford Graduate School of Business, tested this with Claude Code earlier this year. He asked it to update a five-year-old academic paper on universal vote by mail. The tool gathered new data, ran analyses, produced figures and tables, and drafted a new paper — with not very much prompting, Axios reported. Then Hall hired a graduate student to audit the work. The verdict: the agent did a lot right, which is remarkable, but it also made a number of errors. It failed to collect all the data it needed and did not code all the data correctly. It very much needed an expert, PhD-level student to oversee it quite closely.
That is the actual story. Not that AI can do knowledge work — it can, impressively, often — but that the knowledge work it can do still requires a knowledge worker to catch what it gets wrong. The person in the next cubicle is not being replaced today. They are being asked to become a quality controller for a tool they did not build and do not fully understand.
This is the spreadsheet moment. When VisiCalc launched in 1979, the accounting profession did not disappear. It expanded. But the work changed: the junior accountant who spent hours adding columns by hand found themselves managing a model instead of doing the math. The floor rose. The ceiling did not — at least not for everyone. Some people adapted. Some people did not get the memo.
Codex is the VisiCalc moment for coding. The difference is the speed and the scale. OpenAI is not pitching this as a workforce transition tool. It is pitching it as an operating system for knowledge work — a system that connects to your email, calendar, documents, spreadsheets, design apps, Slack, and Teams, and can automate the morning brief you used to write yourself.
The competitive pressure is real. Anthropic launched Claude Cowork in January specifically to capture the non-coder knowledge worker. OpenAI released the Codex desktop app a month later. Both companies are racing to own the desk before the other guy does.
What is missing from this picture is the honest accounting of what happens to the person whose workflow is being automated. OpenAI co-founder Andrej Karpathy, now at Anthropic, described his own experience since December as a "state of AI psychosis" — trying to figure out what was possible and pushing it to the limit, Axios reported. The people being onboarded onto Codex are not being told that the most advanced practitioners of this technology describe it as psychologically disorienting. That might be worth knowing before you set up your morning automation.
The numbers in this story are all OpenAI's own. No independent analyst has verified the five million weekly active users, the three-times growth rate, or the task breakdown. Fortune's March reporting on the 1.6 million baseline is the closest thing to an independent checkpoint — if that figure is accurate, the growth to five million in four months is a real trend, not a rounding error. But it is worth noting: the company that is selling you the future is also the company that wrote the report on how well the future is going.
The spreadsheet analogy has one more thing going for it. Nobody remembers VisiCalc as a crisis. It was a tool. It changed what it meant to be good at your job, and it did it fast enough that most people did not notice the transition until it was already over. Codex is the same transition, running at software speed instead of hardware speed, aimed at a wider range of work, and arriving before anyone has figured out what the quality control function looks like when the model is wrong.
The person in the next cubicle will figure it out. They always do. But the period between when the tool arrives and when the adjustment happens is where careers get quietly revalued — and that period is now.