Should you vibe code a CRM or buy an off-the-shelf one?

"AI writes code so fast now. Why not build a CRM that fits us exactly and skip the yearly subscription?" I have been hearing this question a lot lately.
It is a fair question. Vibe coding, describing what you want in plain language and letting AI write the code, really can produce a working web app in a few hours. You get a customer list, data entry forms and a sales chart, all looking just like the demo of a commercial product.
But what you can see after those few hours is the smallest part of a CRM.
I. WHAT THE DEMO DOES NOT SHOW
A professional CRM is more than an interface. It is also infrastructure running 24/7, data backups, security, permissions down to each user and each record, and a large body of business logic that handles situations you have not thought of yet.
That logic has been built up over many years, from many businesses with all kinds of processes. When you buy an off-the-shelf CRM, you are paying for that far more than for the interface.

II. A SMALL EXAMPLE
I once built a personal expense app. The idea was simple: the app reads transaction notification emails from the bank, pulls out the amount and description, saves them to a database, and draws a monthly spending report.
It was done in one evening. It worked well.
A few weeks later the bank changed its email template. The old way of pulling out the data no longer worked, and the app recorded the wrong figures. At that point there were two options:
Extract the data with code: fast and almost free, but rigid. Each bank has its own email template, and every time it changes, the code has to be fixed.
Let AI read the email and extract the data: much more flexible, it still reads a changed template, but it costs tokens for every email, it is slower, and it sometimes misreads the amount, so the results still have to be checked.
Once that was decided, other small problems appeared one after another:
The bank's promotional emails got in too and were recorded as transactions, so a filter had to be added.
Running the app again recorded a transaction twice, so duplicate protection had to be added.
Refunds, foreign currency transactions and the separators in amounts each needed their own handling.
None of these problems is hard. But they never stop coming, and each one needs a decision.
Then the app ran smoothly and proved useful, so I wanted to share it with a few people to try. That is when the real work started:
The folder paths and file names in the code belonged to one computer, so they all had to be moved into configuration.
The API key sat right in the code, so the app could not be handed to anyone else as it was.
The database was a file on a personal computer. For several users, it had to move to a server.
A server means sign-in, separate data for each user, and backups.
One user opened it in Safari instead of Chrome and the layout broke. Another opened it on Windows and some features did not work.
One personal app, one data source, one user, and it had already come to this.

This is not an isolated case. According to a 2026 study, about 9 in 10 vibe-coded apps in its sample had at least one security vulnerability, including keys left in the code as in the example above.
III. NOW SCALE THAT UP TO A CRM
Replace the personal expense app with a company CRM.
Many users on desktop and mobile, on Windows, macOS and Linux, all working at the same time. The system has to stay fast as the data grows.
Customer data has to be stored safely, and the whole system has to be protected from unauthorized access. Each user needs exactly the right permissions: salespeople see only their own customers, managers see their whole team, accounting sees something else.
A CRM also usually has to integrate with many other systems: accounting, the call center, email marketing, the website, e-commerce marketplaces. With that come API version changes, expired tokens that must be refreshed, call limits, and connection errors that must be retried without creating duplicate data.
All of this is hard and takes a lot of time to design, build and maintain. The vibe-coded CRM suddenly becomes ongoing work, and the company needs someone responsible for maintaining it.
IV. SO WHERE DO YOU DRAW THE LINE
In my view, the line depends on complexity and on the resources the company is willing to commit over the long term.
A simple system is perfectly fine to vibe code: an internal tool for a few users, one data entry screen, one report, no customers' personal data, and something you can rebuild if it breaks.
A complex system needs someone to maintain it for the long term. Without that person, an off-the-shelf system makes more sense. The subscription then pays for not having to carry the infrastructure, security, backups and the problems that come up every week yourself.

Buying off the shelf has its limits too: fees usually grow with the number of users, and very specific needs have to be built the way the system allows.
V. WHAT ABOUT WHEN AI GETS EVEN BETTER
AI may gradually solve many of the difficulties above, and a very small team may be able to build a complex system. But that is not the case today.
Vibe coding is best at producing a system that works. Running that system for many years is a completely different problem.
FREQUENTLY ASKED QUESTIONS
Is building your own CRM cheaper than buying one, once long-term costs are counted?
Maintenance usually costs more than the first build: people to fix bugs, people to deal with broken integrations, infrastructure, backups. Compare the yearly subscription with the yearly cost of the staff who maintain the system, not with the cost of the first build.
Our company has already vibe coded a CRM. What should we do now?
You do not have to abandon it. Check what data it holds, who is responsible when it breaks, whether it is backed up and its data can be exported, and whether anyone has reviewed its security. If any of these is missing, move to an off-the-shelf system early, while there is still little data.
Should you use AI to build extensions around an off-the-shelf CRM?
This is usually a good approach. The off-the-shelf CRM handles infrastructure, permissions and security, while AI quickly builds the parts specific to your business: a custom screen, a custom report, a piece of automation. In the Zoho ecosystem, Zia App Builder in Zoho Creator (a separate product that needs a paid plan) can build a first version with forms, reports and permissions from a description. Approval and payment workflows still have to be set up by hand, and everything AI generates should be checked before use.
Is customer data safer in a cloud CRM than stored in-house?
Usually yes, provided permissions are set up correctly. Large vendors have dedicated security teams, backup processes and regular independent audits. Zoho, for example, holds ISO/IEC 27001 certification and a SOC 2 Type 2 report. A small company would find these hard to rebuild on its own.
Optimus is a Zoho Authorized Partner in Vietnam. Contact contact@optimus.vn for advice on and implementation of Zoho CRM.

Comments