Agentic Data & Software Engineering

Software development outsourcing with AI: the method must be yours, not the vendor’s

By Xavier Franco, CEO de Elternativa · Grupo Zemsania
· 5 min read

Written with the help of AI and reviewed by Xavier Franco

A long wall built in sections, each with a different brick, stone and bonding pattern.
In this article

Few companies build all their software with a single team. The usual setup is a mixed model: an internal team, one or more vendors for projects and another for maintenance. Until recently, the challenge of software development outsourcing was coordinating people. Now there is a new one: each vendor uses AI in their own way. I see it as part of the goal of depending less, and my conclusion is clear: the method has to be yours.

The same thing, done two ways, and both in production

Think about what happens today. One vendor works with an AI tool, with its own instructions and conventions. Another uses a different one, with different rules. Your internal team, a third.

The result shows in the code: the same problem solved in two different ways, and both in production. Different styles, different tests, different documentation. And when something fails, nobody knows what a person wrote and what a model wrote.

In practice: ask each vendor to tell you which AI tools they use on your projects and with which rules. If they can’t say, or each one says something different, you already have your diagnosis.

AI amplifies what is already there

The 2025 DORA report sums it up in a sentence that I find key for outsourcing: AI doesn’t fix a team; it amplifies what is already there. Strong teams use it to get even better. Those with problems watch AI make them bigger.

Applied to a model with several vendors, it means that if each one works in their own way, AI multiplies the differences. More code, faster, and more heterogeneous. And the same report warns that, without solid control systems, a higher volume of changes leads to more instability.

The method can’t belong to the vendor

Until now, each vendor brought its own methodology. It was reasonable: you bought an outcome and they supplied the way to get there. With AI, that has a cost that used to be invisible:

  • The rules live in the vendor’s tools. If you change vendor, they leave with it.
  • What is learned doesn’t come back. It stays with whoever did it, not in the project.
  • You can’t compare. If each one measures in their own way, you don’t know who performs better or at what cost.
  • There is no common traceability. Each vendor records what it wants, where it wants.

That is why I propose reversing the order: you set the method, and the vendors work within it.

In practice: in your next development contract, include a clause on where the rules, decisions and knowledge that are generated end up. Make it a place of your own.

What has to be common

A common method doesn’t mean everyone uses the same tool. Each vendor can use the AI they prefer. What has to be the same for everyone is this:

  1. How a task is born: with its contract and acceptance criteria fixed before starting.
  2. Who tests: a different instance from the one that wrote the code.
  3. Who validates: a person on your side, who looks at the code and signs off.
  4. Where what is learned is kept: rules, decisions and mistakes, written in a system of your own.
  5. What is measured: the same metrics for everyone, taken from your Git.

With that, vendors keep contributing their people and their experience, but the result is homogeneous, traceable and comparable.

The rules, by levels

In practice, the rules are organised by levels. What comes built in with the platform; what is common to your company, such as your AI tools and conventions; what belongs to each client or department, such as its delivery policy; and what belongs to each project, such as its branch and environments.

What you define at the top applies to everything below, and what is below refines it, never contradicts it. That way, a new vendor joining a project receives from day one the same rules as the previous one.

In practice: write down on one page the five conventions you want anyone developing for you to follow. That is the first level of your method.

Changing vendor without losing anything

The most important effect is this: changing vendor stops being a risk. If the method, the rules, the decisions and the project knowledge live in a database of your own, the incoming vendor starts where the outgoing one left off.

And it opens one more option: setting up your own Technical Office, in which your teams and your vendors work with the same method, each in their own role.

Where to start

  1. Ask each vendor which AI they use and with which rules.
  2. Define the first level of your method: what is common to anyone developing for you.
  3. Decide where knowledge lives, in a database that is yours.
  4. Measure everyone the same way, with metrics from your Git.
  5. Test the method with one vendor and a limited scope, and compare.

How we help

At Agentic Data & Software Engineering we orchestrate development AI under a single method for everyone: contract and criteria before starting, tests by a different instance and validation by a person. We run it in the engagement model you already buy, as technical assistance, a fixed-scope project or a managed service, and, if you like, we help you set up your Technical Office so that your teams and vendors can operate it.

We do it with Agentagile: the method lives in the system, by levels, and the roadmap, contracts, decisions, traces and project knowledge are kept in a database that you own.

Frequently asked questions

How does AI affect software development outsourcing?

Each vendor uses it with their own tools and rules, and that multiplies the differences in code, tests and documentation if there is no shared method.

Do all vendors have to use the same AI tool?

No. What has to be shared is the method: how a task is born, who tests, who validates, where what is learned is kept and what is measured.

How do you avoid losing knowledge when you change vendors?

By keeping the project’s rules, decisions and documentation in a database owned by the company, not in the vendor’s tools.

What is a Technical Office with AI?

A way of working in which your teams and your vendors develop with AI under a single method, operated by your organisation.

Keep reading