On 5 November 2026, OpenAI published a piece on Cult.fit. Teams across the company, not only the engineers, are using Codex to build their own tools. OpenAI’s summary says those teams move faster and that it has changed how work gets done there.
The claim, and the test I apply
The excitement around a story like this is that anyone can build software now, so every team becomes a software team. My first question is always whether there is a real need. Here there is.
Most businesses carry a layer of work that lives in spreadsheets, exports and one person’s head. Someone pulls three reports every Monday and merges them by hand. Someone keeps the roster in a sheet with colour codes only they understand. Someone checks stock levels against last month’s sales before placing an order. None of it is hard. All of it eats hours, and the person doing it is usually the one who knows the process best.
What changed is that this person can now build the tool. Before, they wrote a request, a developer queued it behind customer-facing work, and by the time it shipped the brief had lost half its context. Now the distance between the idea and a working tool is small. That is the only distance vibe coding collapses, but for this kind of work it is the distance that mattered.
What did not change is everything else. Real-world testing, user feedback, security, and the back-and-forth on what the tool should actually do. No tool knows your refund limit or your delivery cut-off until someone with that context tells it, and you do not get that from a single prompt.
Back-office tools are personal tools with more than one user
I have said before that personal tools are where the vibe coding shift is real. A normal person can build fairly competent software for their own use, and the fallout is small because it only handles their own data. I replace many of my own tools and scripts this way.
An internal utility for a team of three has the same shape. It reads the company’s own data. The people who use it sit next to the person who built it. If it breaks on Tuesday, the builder notices on Tuesday and fixes it. Nobody outside the business ever sees it. That is why Cult.fit’s story reads to me as the personal-tool shift arriving inside a company, not as a new category.
B2B software still has a gap that vibe coding cannot close, and that gap starts exactly where the tool stops reading your data and starts acting on someone else’s.
Where the line sits for an operator
Take a clinic. The front desk builds a small tool that turns the appointment export into a list of patients to chase for no-shows. Build that. Taking the deposit, sending the reminder on WhatsApp and moving the booking when the patient replies, buy that and integrate it.
Take a shop. A reorder check that reads the stock export and flags what runs out next week. Build it. Booking the courier, issuing the refund, telling the customer where the parcel is, buy it.
Take a marketing team. A reporting tool that merges ad spend, search data and orders into one weekly view. Build it. Anything that sends messages to customers at scale, buy it.
The rule I use is simple. If the tool only reads your own data and only your team uses it, build it yourself. If it commits money, promises a date, talks to a customer or moves a parcel, buy it and integrate it, and keep the rules about what it can commit to in the system rather than in a prompt.
I see this from the other side with AeroChat. It connects to each market’s payment processors and logistics partners, and those relationships took real work to set up. You can’t vibe code your way into them. That is the difference between a tool and a product.
What a business should do now
- List every spreadsheet, export and manual handoff your team runs each week. Most businesses find more than they expected.
- Let the person who runs each process build its replacement. They carry the context. Do not route it through a developer queue.
- Keep those tools on your own data, used by your own people. That keeps the fallout small when something is wrong.
- For anything that touches a customer, a payment or a delivery, buy it and integrate it. Put the limits in the system, not in the instructions.
- Check every output before it leaves your hands. You direct, the AI produces, you check and decide.
Source: OpenAI News: Cult.fit builds operational muscle with Codex