Yves Caseau, Chief Digital Officer at Michelin, offers a nuanced reflection on the evolution of IT roles in the face of the emergence of generative code AI. Far from predicting the obsolescence of developers, he envisions a profound but structured transformation of roles.
AI transforms software architecture
First, generative AI will transform software architecture by enabling orchestration agentique. Future systems will simultaneously serve humans and autonomous agents. However, this evolution does not eliminate programming: it simply changes its modalities. The traditional user interfaces of SaaS applications will be lightened or disappear, replaced by flexible backend services. Nevertheless, programming remains a complex activity requiring rigor and logic, regardless of the language used—whether Python, Java, or natural language.
Three categories of professionals emerge
Second, systems engineering retains vital importance. Three categories of professionals will emerge clearly. Computer System Engineers will continue to build foundational architectures in traditional AI-assisted code, managing the critical backbones of systems. Solutions Engineers will program primarily in natural language to orchestrate agents and compose complex solutions. Citizen developers will gain access to sophisticated systems without advanced IT skills, democratizing application creation.
At the same time, user interface tasks will face three distinct fates: simple tasks will disappear, absorbed by agentic intermediation; moderately complex tasks will migrate to accessible no-code tools; only truly complex cases will remain with professional IT staff.
Context remains king
Third, writing code becomes relatively trivial compared to software maintenance. System adaptation depends deeply on context: the origins of changes (users, business processes, technical environment), the impact of legacy systems, intrinsic business complexity. The transition to the agentic approach is generally overestimated in discourse, while the inertia of legacy systems is systematically underestimated.
Caseau astutely points out that every historical level of abstraction has reached its limits—assembler to C, C to high-level languages, procedural to object-oriented. Natural language, while powerful for orchestration, does not always constitute an optimal specification language. Sometimes, precisely specifying a system requires as much cognitive work as programming it directly.
A balanced vision
This balanced vision acknowledges the transformative impact of AI while preserving the fundamentals of software engineering: complexity management, robust architecture, and deep understanding of business context remain inalienable skills. Abstraction is indeed increasing, but responsibility and architectural coherence remain central—AI accelerates production but eliminates neither complexity nor the need for deep architectural expertise.