Before you put AI into your SaaS, take a moment.
A practical approach to adopting AI agents without compromising security, quality, or accountability.
I have to admit it: when I started seeing people in my work environment using agents on their own (marketing, sales, HR…), I was surprised. Before that, I thought:
We developers will be the first to use this!
And it wasn’t like that.
That teaches us something fundamental:
👉🏼 AI adoption in a company isn’t just technical, it’s cultural.
And if you want it to be safe and useful for your SaaS product, you need onboarding.
Below is the minimum workflow we proposed and started implementing in the teams I work with, based solely on our experience.
But first things first.
Why do you need formal onboarding?
Because without it:
People use agents without assessing their security or legality.
You can end up exposing sensitive data or violating standards/ISO regulations.
Teams create silos of “how” to use AI, and no one coordinates.
In my case, we had to stop and think:
How do we do this without messing things up?
So here’s my proposal, based on our teams’ experiences.
My proposal
1️⃣ Assessment
What uses already exist in the Company? Marketing and sales were already using tools to create content; developers were testing Cursor/Copilot.
The process team handled this analysis and conducted surveys to understand the needs of each department.
2️⃣ Minimum training for everyone
Not just developers. We provided training for product, sales, and HR: what generative marketing is, legal limitations, and best practices.
This was done in parallel with the assessment mentioned above.
3️⃣ Choosing tools and understanding how to use them
In my case, we opted for a ChatGPT-type system but designed for internal use within a company (globally). We chose Nexus AI.
For development, we discussed Cursor, Windsurf, and GitHub Copilot, which also included training courses for its tools.
4️⃣ Adoption
No adoption, no party. If you don’t achieve adoption of a system, process, or tool, it’s as if you haven’t done the work.
This was communicated repeatedly via email and in global meetings. etc., regarding the existence of the different tools available to employees in general and developers in particular.
As you can see, it’s not a particularly sophisticated onboarding process. When we consider adopting AI to improve our productivity, we don’t find onboarding experiences from other companies.
Everything moves very quickly, so we did what we’ve done before, but geared towards AI.
✨ Pro Tip for Developers
When starting to work with an AI code assistant, my recommendations are as follows:
Write your prompt as if you were explaining to a junior developer (or another junior developer if you’re a junior developer) what you want to achieve.
Give them instructions on how you want the code to be structured. I’m referring to things like software architecture, best practices, etc.
Write the tests yourself and ask the AI to write the code based on the tests you’ve written. As a developer, you know the use cases, so write the tests and let the AI do the work. Initial implementation. For this, I recommend you read this email: Workflow with TDD and AI Agents.
And always review the code that the AI generates.
☝🏼 Start with tests and ask the agent to implement those tests. You remain responsible; the AI accelerates the process, but don’t delegate the responsibility.
Did you like this edition? Share it with someone who’s working on this 👇🏻