When AI is integrated into every stage of development, the developer shifts into a role of managing the big picture.
Stefano Garusi has been working in software development for 20 years, ten of which as a consultant, so the changes in the industry are nothing new to him. However, AI-driven and agentic development have introduced so many new possibilities that when we interviewed him about the transformation of his work, Stefano ended up sharing a far-reaching vision for the future of developers and the software industry. In his view, these changes extend from automated development pipelines to how organizations develop, learn, and communicate.
Note on the videos: despite the thumbnails, the interview was conducted fully in English.
Software development has already changed, but not nearly as much as it´s about to.
According to Stefano, almost all organizations are currently striving to rebuild their development cycles with an AI-first approach. He believes that AI can be utilized at every stage of the development process, from requirements definition to design, implementation, testing, and maintenance.
Organizations are currently engaged in extensive planning and testing to define how AI tools can be best utilized at each stage. After this, it is also crucial to chain these stages together so that the output of one stage becomes the prompt for the next. Only by planning the development cycle in this way can we begin to move toward automation that covers the entire development pipeline.
The value of an experienced developer is largely in their head. It lies in the answers to questions like, "How do these parts of the system relate to each other?", "Why was this feature built this way?", or "What does the customer actually mean and want?" Stefano believes this tacit knowledge must now be captured and stored for AI to use.
When I develop features, I have all the connections in my head. Now, all those connections and that context need to be transferred to the AI.
Current developers must learn to identify their own thought processes and write them out. Documentation needs to become more precise, and requirements must be formulated so unambiguously that AI can act on them effectively and accurately. None of this will be achieved alone; developers will need to collaborate more than ever within their organizations.
This process doesn't just have to be for the benefit of AI and agents. Once information is written down, it is available to the entire organization.
Stefano emphasizes the importance of data collection, whether through documentation or automation: "Whenever you develop features, you have conversations with different team members—and nowadays with AI—and you learn new things. Now, this same learning process should be transferred to the AI."
User stories, documentation, and conversations in Teams and Slack are currently scattered in different places. In Stefano's view, these will form a unified context in the future from which AI and agents will draw. It will become the developers' job to build the methods by which the accumulation of this collective memory is managed, guided, and automated.
An AI that learns and remembers doesn't just have to be an aid for software development; the idea can be applied to the entire organization's operations. Stefano admits that this is still a high-level concept in his mind, but notes: "Especially in larger companies, this could help the entire organization learn."
I believe AI can help answer the question: "How does our organization learn?"
By collecting enough data from various sources and in the right format, Stefano believes an organization can form a shared knowledge base—a kind of central brain that models what is happening within the organization. Anyone can use it to get answers to their questions, and the information becomes more precise every day through new interactions.
"In my experience, differences in perspective between teams are the biggest source of errors, misunderstandings, meetings, and wasted time." Stefano feels that AI can also act as an interpreter between different teams, thereby saving everyone's time.
A software professional sees things from the perspective of software development. They may have some idea of how sales sees the same issue, but not necessarily the words to describe their own view in the language of sales. Stefano believes that a language model fed with all perspectives and the discussions surrounding them is capable of doing just that.
The end result is smoother communication within the organization, which can also lead to long-term cost savings for the company as time management becomes more efficient.
Stefano is enthusiastic about the future, but reminds us that you get the most out of AI when you do your groundwork carefully. This is because AI is ultimately an amplifier of processes: a well-defined process multiplies quality, while a poorly defined one multiplies errors. That is why it is worth investing in the development pipeline, guidelines, and quality assurance right from the start.
If there is an error in an agentic development pipeline, it will also multiply those errors.
In Stefano's view, the biggest open question in the field, and at the same time the greatest opportunity, is figuring out what kind of prompt produces the desired result. “AI is a fantastic tool, but we are still talking to a computer,” Stefano points out, emphasizing that it is important to spend time testing, refining, and developing prompts to ensure the process is reliable enough. This is especially critical when developing the entire development pipeline, as the quality of the prompts must remain high at every stage of the process.
For Stefano, the change brought by AI has been a welcome one. “I enjoy coding, but I have always been more interested in the business outcomes of software development.”
Stefano believes that the developer's role is increasingly shifting to overseeing the software development process. In the future, the developer will act as an interface between the business, the customer, and quality assurance. They will work with everyone to clarify what is being built and articulate the requirements so precisely that the AI can execute them.
In the future, we will oversee the process from start to finish much more comprehensively than we do now.
At the same time, the developer ensures that the information flowing between stages is accurate and precise. Understanding business and grasping the big picture have always been important strengths for a consultant.
Now, they are moving to the very core of their work.
