Buy or build: the freedom AI gives back to us
By Xavier Franco, CEO de Elternativa · Grupo Zemsania
· 7 min read
Written with the help of AI and reviewed by Xavier Franco

In this article
Rarely have I felt in my work what I feel now. AI has given us something that until very recently was unthinkable: being able to dream up a tool and see it working within weeks. That is building. And for a company it means getting back a freedom we had lost: the freedom to choose what we buy and what we build.
The decision we had put off for years
For two decades, the almost automatic answer has been to buy. An ERP, a CRM, a BI tool, a SaaS for every department. It made sense: custom development required a large team of analysts, developers and testers, months of project work and maintenance that nobody wanted to inherit.
We all know the price of that decision:
- We adapt the business to the software, not the other way round.
- We wait for the vendor for the feature that sets us apart, and sometimes it never arrives.
- We pay for what we do not use and build patches around what we do.
Buying was not always the best option. It was the only viable one.
In practice: list what you solve today with spreadsheets, manual processes or «when the vendor ships it». That is where what you never considered building lives.
What has changed with AI software development
The cost of building has changed. According to Google's 2025 DORA report, 90% of technology professionals already use AI in their work and more than 80% say it has made them more productive. In the 2025 Stack Overflow survey, 84% of developers use or plan to use AI tools.
But the same DORA report says something I find the most important: AI amplifies what is already there. Teams with good practices go faster; those without them see their problems grow. And trust does not keep pace: 46% of the developers in the Stack Overflow survey distrust the accuracy of what AI generates.
That is the key. AI does not give us speed alone. It gives us the possibility of building with guarantees, if we work with a method. Without one, we only multiply the mess.
In practice: before asking which AI tool to use, ask yourself whether your way of developing would withstand twice as many changes per week.
Buy or build: it is not going back
I know what many of you are thinking: «we already went through custom development and it was expensive». I am not proposing we go back to that. I am proposing we choose again, case by case, under different rules.
Keep buying when:
- the process is standard in your sector and does not set you apart (payroll, accounting, email);
- there is a regulation behind it that the vendor keeps up to date for you;
- the product already covers what you need without patches.
Consider building when:
- the feature is the one that sets you apart from your competitors;
- you have spent years waiting for the vendor or working around the product with spreadsheets;
- you pay for licences while using only a small part of them;
- you need to connect systems that do not talk to each other.
And often the answer is mixed: you keep the core you bought and build around it whatever makes you different. That is not going backwards. It is going faster and being more agile.
In practice: put each case on your list in one of the two columns. If in doubt, the question is simple: does this set us apart or does it just keep us running?
What changes for each person in the company
This freedom is not just for IT. It touches several desks, and each one should know what it gains and what it has to contribute.
- Management: stops choosing only between vendors and goes back to deciding which capabilities it wants to own. The question changes from «what do we buy» to «what makes us different and should be ours».
- Business: gets its voice back. Whoever knows the process can describe what is needed and see a first version working, without waiting for the next purchasing cycle.
- IT: moves from integrating other people's products to governing what is built: what comes in, with which tests and who validates it. Its role grows, it does not disappear.
- Procurement and finance: gain a real alternative at every renewal. You do not need to break with the vendor to negotiate better; it is enough to know that building is an option.
None of these changes requires hiring an army of developers. They require agreeing on the method and on who decides.
In practice: at the next renewal of an important licence, ask IT and the business for an estimate of what it would cost to build the part you actually use. Even if you renew in the end, you will negotiate differently.
Building with guarantees and without a big team
This is the part that excites me most. You no longer need a huge team to build well. You need a method that turns the AI you already have into a real team. In AI software development, the guarantees do not come from the tool. They come from how you work:
- Criteria before starting. Every task begins with what it has to meet to be accepted, and that does not change afterwards.
- Independent tests. Whoever writes the code is not the one who tests it.
- A person decides. Nothing goes to production unless someone on your team validates it by looking at the code.
- Everything in-house. The code in your Git; the project knowledge in a system of your own, not in one person's memory.
DORA's AI capabilities model points in the same direction: among the practices that most multiply the effect of AI are working in small batches, solid version control, a focus on the user and quality internal platforms. None of that is new. What is new is that it now makes the difference between building with guarantees and building in a hurry.
With this, your team stops hammering out code and moves to what really adds value: defining what the business wants and validating what is delivered.
In practice: look at your last project. Who decided the acceptance criteria, and when? If it was at the end, that is the first change.
Where to start
My advice is to start small and measurable:
- Pick one case from the «build» column: well bounded, useful and with someone in the business who really wants it.
- Write what it has to do and how you will know it works, before writing a single line.
- Put the method before the tool: contracts, separate tests and validation by a person.
- Measure on your own repository: deliveries, rework and time to production.
- Decide with data whether to scale to the next feature.
In a few weeks you will know whether the freedom to choose is real in your company. I am betting it is.
How we help you
At Agentic Data & Software Engineering we orchestrate the AI development platforms you already have, such as Claude Code, Cursor or Codex, so that they work in parallel, unattended and governed, across the whole software development lifecycle. You do not scale people: you scale deliveries. And what did not fit before now fits in the plan.
We do it with Agentagile, our methodology for developing software with AI. Your AI develops, tests and documents. Your team defines and validates. Every delivery is tested and validated under the same rules, and the knowledge stays in-house.
Which feature have you been waiting years for from your vendor?
In a demo, we show you how the AI you already have develops it, tests it and delivers it, with your team validating every step.


