AI does not replace engineers but reinforces the need for collaboration and knowledge sharing to form profiles.
Every discussion about AI in software development always comes down to the same assumption: equipping engineers with smarter AI tools would mean that a single engineer could do the work of an entire team. It’s an attractive idea, but a misguided one.
AI has expanded the scope of what an individual developer can do, but it has also expanded the scope of knowledge expected of that developer. An engineer who relies on AI to produce infrastructure code must still evaluate its security implications. Even with an AI-assisted security scan, one still needs to understand the underlying business logic. The more tasks AI takes on, the greater the judgment required to validate what it produces.
The organizations that will get the most out of their investment in AI will be those that deliberately invest in collaborative practices, cross-functional reviews, structured knowledge sharing, and targeted mentoring, so that every engineer gains the multi-domain mastery that AI needs but cannot provide. Let us cite the example of Radio Francewhich adopted GitLab to unify development and deployment workflows across teams. By moving to a single platform, teams reduced context switching and gained shared responsibility for delivery. Deployments became 5x faster, cycle time decreased by 82%, and teams realized substantial savings.
Great software doesn’t just result from better tools. It results in better teams.
Collaborative foundations as a solid foundation
The central goal of DevSecOps is to establish a collaborative engineering culture spanning the entire software lifecycle, from business strategy to technical implementation. This culture emphasizes reuse and best practices, which directly improve developer productivity and delivery efficiency.
Organizations achieve this through a dual control system:
- Code reviews based on human consensus ensure knowledge transfer and maintain quality standards across disciplines.
- Automated quality and safety checks detect problems before they reach production.
This approach balances speed and control. It minimizes the risks associated with software change management while ensuring that acceleration does not come at the expense of stability or security.
Most organizations implement the processes, install the tools and measure the velocity gains. Yet they miss the deeper transformation taking place behind the scenes.
Knowledge transfer mechanisms
The collaborative model fundamentally functions as a large-scale learning and knowledge acquisition system. Research in educational psychology, in particular Bloom’s taxonomysuggest that the highest form of understanding is achieved by teaching concepts to others.
This is where the double door system reveals its deeper value. Code reviews become structured knowledge transfer sessions. Each person acts as an expert in their field while learning adjacent areas:
- The security engineer who reviews code teaches secure development practices while becoming familiar with business requirements,
- The architect understands the priorities of the product while sharing his knowledge on technical constraints,
- The junior developer learns the models from the seniors while bringing a fresh perspective to the tools.
This creates a network effect where everyone’s knowledge elevates everyone’s capabilities. Expertise flows in all directions. This collaborative culture transforms the organization into a learning organization, where each interaction becomes an opportunity for teaching and accelerated progress.
From this perspective, each code review becomes an educational moment, each security scan a learning opportunity. This is what sets some engineers apart: they have internalized knowledge from adjacent fields through years of collaborative interactions.
The autonomous engineer: AI as a partner, not a substitute
The natural evolution of this collaborative model is the “autonomous engineer,” an AI-augmented knowledge worker who enables unprecedented independence and efficiency. The promise remains attractive. Each engineer has AI partners who handle low-level tasks, such as memorizing, understanding, and basic application of concepts. Assigning these redundant tasks to an agent significantly reduces cognitive load, freeing up mental capacities for higher-level thinking, including analysis, evaluation, and creative problem solving.
This is how AI can amplify human capabilities rather than replacing them. A recent GitLab study found that while 83% of DevSecOps professionals believe AI will significantly change their role over the next five years, 76% agree that AI will actually create an increased need for engineers, not the other way around.
However, a dangerous counter-narrative is emerging in ruling circles. Some executives believe that high-performance AI agents can replace knowledge workers entirely. This represents a fundamental misunderstanding of how people develop their expertise.
Even with high-performance AI, you still need human experts who can:
- Evaluate results in several disciplines,
- Establish trust in AI recommendations,
- Provide domain-specific judgment,
- Take responsibility for production systems.
In fact, a GitLab study found that 40% of DevSecOps professionals agree that AI tools will actually accelerate the career progression of junior developers.
The argument that “we don’t need junior developers anymore” ignores the fact that someone still needs to review, validate, and take responsibility for what AI produces. Junior developers don’t just write code; they learn to evaluate it in multiple domains, thereby gaining the judgment necessary to verify AI results.
The opposite argument, that AI could replace experienced architects and senior developers, remains just as problematic. This logic suggests that we could ignore foundational learning altogether and restructure computer science education to focus solely on programming AI agents. But without understanding what good code looks like in security, infrastructure, and business, how would these graduates know if the AI results are correct? These two extremes miss the point.
The real bottleneck: limited collective wisdom
The real constraint is not the capacity of the AI. This is the rarity of people capable of acting as an “autonomous engineer”. Engineers with sufficient skills across multiple domains are required to effectively evaluate AI outcomes in security, infrastructure, quality, and business logic. And we need trainers who understand how to train these versatile professionals.
The collaborative model from the original goal of DevSecOps remains essential, as it is the mechanism by which individuals develop the breadth of their knowledge. The autonomous engineer is not someone who works in isolation. This person has internalized the collective wisdom of the cross-functional team and can now work with the help of AI while retaining the judgment and accountability that only human expertise can provide.
The way forward
Organizations face a crucial choice. The tempting path is to view AI as a cost-cutting strategy by replacing expensive senior talent with cheaper tools and anyone who knows how to use them. This path leads to fragile systems, technical debt, and ultimately failure.
The sustainable path recognizes that AI serves as a tool that amplifies existing capabilities, but cannot replace the judgment that comes from deep, cross-functional knowledge.
The companies that will prevail are those that focus more on collaborative learning while simultaneously investing in AI augmentation. They understand that to create an autonomous engineer, you must first build a team that trains each individual in multiple areas. They recognize that the code review process transfers the knowledge needed to use AI tools effectively. They invest in setting up knowledge transfer systems that train engineers capable of working independently, after having learned from the collective.
This illustrates the paradox of the AI era in software delivery. As our AI tools become more and more capable, the value of collaborative learning becomes essential. The only way to train individuals capable of exploiting these tools is through the interdisciplinary knowledge transfer enabled by DevSecOps.
We must always increase productivity, maximize efficiency and reduce risk. What has changed is our understanding that achieving these goals at scale requires both collaborative learning and AI augmentation, not a choice between the two.
The future belongs to organizations that develop cultures where everyone teaches, everyone learns, and everyone becomes able to function as an autonomous engineer when augmented by AI. Ultimately, the real competitive advantage is not AI; these are the people who know how to apply it effectively.
