I sketched the first version of this framework to clarify for myself where I stood, but also to give direction to the Blis Digital team. The result was our AI adoption framework: four levels that show how you, as a software professional — or team — grow towards AI-first work. We are now using it at Blis Digital and with customers. And although you should never follow a framework too dogmatically, it provides a lot of guidance.
Level 1 — AI User
At this level, you're using AI as a tool, but it doesn't change how you work. You use GitHub Copilot or ChatGPT for a piece of code or to summarize a text. But your workflow will continue as it was. AI is being used passively. That's how I started, too. A bit of prompting. See what comes out. But without knowing why it sometimes works, and sometimes not at all.
And that's where the risk lies: you think you're using AI, but in reality, your work process remains exactly the same. Developers still rely on their own routines, but with some extra suggestions. The difference in output is small, so is the impact on the overall work process.
Level 2 — AI Collaborator
The real change came when I started seeing AI as a colleague. An AI that lets you try something, gives feedback, keeps you on your toes. Then you are actively collaborating with AI. For example, you let him create a first version of a function, but you decide the approach, the preconditions, and you actively control what comes out.
At this stage, you'll also learn how to write better prompts and really incorporate AI into your work process. For example, our people learn to let the AI first generate a solution direction — including thinking steps. Only then do they ask for code. This prevents a lot of frustration. Another important part of Level 2 is that you learn to switch between trust and control. You know when to use AI input, and when it's better to get started yourself. It's the start of an “AI-aware mindset” — you don't see AI as a trick, but as a structural part of your work.
Level 3 — AI Orchestrator
At some point, I got to the point where I was going to outsource work to AI agents. No more simple prompts, but whole parts of a project that I handed over. An example: I had screenshots of an old existing application, the data model and our guidelines for a modern cloud-native application. I gave it to the AI and asked: build the functionality. What came out? 80 to 90% usable code. In one afternoon.
This level is about the smart use of AI in entire workflows. It's not just more collaborating, it's really orchestrating. You design a task for the AI, including context and boundaries, get the work done, and then assess the output as if it came from a junior developer. At this level, we train people in test-driven development with AI, among other things. That means: let the tests be written first, then the code. You define success criteria before implementation. This allows you to control AI in a much more targeted way and better control what comes out.
Of course, QA professionals can also operate at this level. One of our QA leads uses an LLM agent as an automated tester. This agent simulates user behavior, performs variations, and reports anomalies. But human control remains essential: someone must determine what is really relevant and what is not.
Level 4 — AI Champion
The step to Level 4 is not only technical, but also cultural. You help others grow in their AI skills. You give workshops, you let colleagues watch, you experiment with new tools — and share what works and what doesn't. At Blis, we have formed a leading group of people who get energy from it. Who are curious. Who had already started with new tools before we rolled out anything. We give them the space to pioneer.
At Level 4, there are people who not only use AI themselves, but take their entire team with them. They create documentation, design internal processes, help with governance and tooling choices. They ensure that AI is not something that a few experts do, but something that benefits the entire team.
Common language
The framework gives us a common language. It helps with staff planning, coaching, and forming project teams. And it shows what we've often felt: that working AI-first is not an all-or-nothing story, but a development path.
We now also use it for resource planning. If a team mainly has Level 1 and 2 people, we know that Level 3 or 4 needs to be added to make a real impact. And the opposite also applies: we encourage experienced people to coach other team members so that they don't have to take all the AI initiatives themselves.
Start by determining where you stand
Want to start your journey to AI-first work? My advice: start by determining where you stand. Ask your team where they see themselves. And let the framework not be a yardstick, but an invitation to grow and learn. Most organizations underestimate how many people in their team can already do — and overestimate how much changes naturally.
A clear structure helps. But it's only valuable if you give people the space and responsibility to grow in them.
This is part 2 in a five-part series about working AI-first in software development. In the white paper “The Foundation of an AI-First Company”, you can read about the framework we used to make Blis Digital AI-first and what we're currently using to make our customers AI-first.
Read part 1: Why many software companies are in danger of missing out on the AI boat
.jpg)
%20veiliogheidsbeeld%20klein.jpg)