For the past year, one question has appeared in almost every discussion about the future of software development: will AI replace developers?
At Nion, our answer has been consistent. No, we don’t believe AI will replace developers. But we do believe developers who fail to learn how to work effectively with AI risk being left behind.
The reason is simple. Software development has always evolved alongside new tools. Version control, cloud platforms, automated testing, CI/CD pipelines, infrastructure as code, each changed the way teams work. AI is the latest shift, but it may be the most significant one yet.
The advantage won’t come from simply having access to AI, but it will come from knowing how to use it well.
We also knew that successful AI adoption would require more than technical skills. It would require clear governance, approved tools, security guardrails, and a shared understanding of how AI should be used in client work. From the start, our goal was not simply to introduce AI, but to introduce it responsibly and consistently across the organisation.
That belief is why we built an internal AI Adoption Strategy, open to the whole company, taking us from what intelligence actually means to what it feels like to build and deliver products with AI as part of the development process.
The Four Phases of AI Adoption
We knew from the start that adopting AI across the company would require more than simply purchasing licences and hoping people would use them. Instead, we approached it as a structured learning journey built around four phases.
Phase 1: Introduction to AI
The first step was creating a common understanding of the technology. Before discussing tools, prompts, or use cases, we wanted everyone to understand what AI is, how it works at a high level, and why it matters. The goal was to build curiosity while cutting through both the hype and the fear surrounding AI.
Mikael Rickan opened our AI journey with The Foundations, aligning the whole company on a shared understanding of AI, intelligence, opportunities, and risks before anything technical entered the room.
Phase 2: Choosing the Right Tool for You
Different roles use AI differently. A developer, consultant, project manager, marketer, or salesperson may solve completely different problems with AI. Rather than selecting a company standard from the outset, we gave employees access to a limited number of approved AI tools that met our security and compliance requirements and encouraged them to evaluate those tools in their daily work.
This phase was deliberately hands-on. Employees were asked to select a tool and actively use it in real work rather than relying on demos or isolated experiments. The objective was not to identify a universally “best” AI tool, but to understand which tools delivered the most value across different roles, teams, and use cases.
A key part of this phase was self-testing. Employees were encouraged to challenge their chosen tools with increasingly complex tasks, refine their prompts, provide better context, experiment with different approaches, and reflect on the quality of the outputs. In practice, this became a lesson in AI literacy. People quickly discovered that better prompts, clearer instructions, and stronger domain knowledge consistently produced better results.
By the end of the phase, we had practical experience, shared lessons, and a growing understanding of how different AI tools performed in real work environments.
Phase 3: Understanding How It Works
Once people were actively using AI, we shifted focus to understanding what is happening behind the scenes. We explored topics such as large language models, embeddings, agents, and real-world AI applications. Our belief was simple: the better people understand the technology, the better decisions they can make when using it.
Filip Stefanovski spoke about How LLMs Work, giving our technical consultants a grounding in what actually happens under the hood of the models we use daily. Stefan Aleksikj’s Beyond the Chat Box traced the shift from chat window to co-worker, showing how these tools are evolving from answering questions to solving problems, interacting with systems, and carrying context between tasks.
Understanding the technology helped move the discussion beyond prompts and productivity hacks, giving employees a clearer picture of both the capabilities and limitations of modern AI systems.
Phase 4: Share Your Knowledge
The final phase was turning individual learning into company-wide learning. Employees were encouraged to share what they had discovered, what worked, what didn’t, and how AI was affecting their daily work. This helped us build collective knowledge faster and avoid every team solving the same problems independently.
Filip returned for a hands-on session on text embeddings and AI agents, building the practical foundation for what came next. Gabriel Wärmby delivered one of the strongest proof points of the programme: a working application built and shipped in a single month using a multi-agent setup, despite not being a developer himself. The season closed with Ivana Prosheva Gligorovska’s session on AI brain fry and burnout, addressing the cognitive impact of heavy AI use and why IT consultants should be aware of it.
An Outcome Beyond a Training Programme
Our journey so far has taken Nion from a shared definition of intelligence to a real, shipped product built with AI agents, without skipping the human cost along the way.
But the outcome is bigger than a training programme.
What started as a learning initiative has become part of a broader organisational effort to integrate AI into the way we work. Across the company, teams are testing AI in real projects, sharing experiences, challenging assumptions, and helping shape best practices for responsible adoption. Alongside new skills, we are building governance, security practices, approved tooling, and shared expectations around responsible use.
The goal was to help the entire organisation develop the capability to work effectively with AI. Not everyone will use AI in the same way, but everyone should understand its potential, its limitations, and how to apply it responsibly in their role.
Technology alone does not create a competitive advantage. Sustainable advantage comes from people, processes, and the ability to adopt new technology in a deliberate and consistent way. By investing in education, experimentation, governance, and knowledge sharing, we are creating a foundation that allows AI to become part of how we work rather than a collection of isolated individual initiatives.
Our investment isn’t only about AI tools, but about the skills, practices, and organisational capabilities that will help us continue delivering value as our industry evolves.
Choosing a Standardised AI Platform
To conclude, the experience gained during this process also informed our decision on which AI platform to standardise on across the company. You might have seen a hint if you’ve been following us 😊
During the summer, a group of colleagues completed Anthropic’s official learning path as part of our preparation for certification. The programme covered four areas:
- Introduction to Agent Skills: building, configuring, and sharing reusable skills in Claude Code.
- Building with the Claude API: working with Anthropic models through the Claude API and understanding how to integrate them into applications.
- Introduction to Model Context Protocol (MCP): learning how to connect AI models to external services through tools, resources, and prompts.
- Claude Code in Action: integrating AI-powered development workflows into everyday software delivery.
The next step is completing the certification exam. While the certifications themselves are valuable, the larger objective is developing hands-on experience with the technologies and practices that are increasingly shaping modern software development.
Stay tuned for more updates on our company-wide AI adoption journey.