PLM and AI: two timeframes, one trajectory
Driven by generative AI, PLM is gradually evolving from a static repository into an active design support engine. In the aerospace sector, this transformation is unfolding in stages: the automation of simple tasks is already delivering its first benefits, whilst the shift towards agent-based systems is being planned for current and future aircraft programmes.
The aerospace industry’s PLM (Product Lifecycle Management) system encapsulates a unique technical heritage: decades of configurations, quality feedback and engineering decisions. The emergence of artificial intelligence, and particularly generative AI, promises to unlock this dormant value. However, the sector is not making the transition at the same pace as others: its systems are critical, its cycles are long and its margins for error are narrow. Damien Constantin, Director of PLM Operations at Sopra Steria, and Mathieu Mollin, Technical and Innovation Director for the Aeroline vertical at Sopra Steria, observe a two-speed movement: automation that is already here, and a more profound transformation driven by agent-based artificial intelligence.
Automation is already at work
The most widespread use of AI in PLM currently involves the automation of simple, low-value-added tasks. “This frees up time for design engineers to focus on higher-value activities, but it is not yet bringing about a fundamental change in working practices,” explains Mathieu Mollin. Damien Constantin agrees: AI speeds up data navigation, facilitates impact analyses and streamlines designers’ work. PLM solution providers are now integrating these features natively into their solutions, making them accessible as soon as they are deployed in a customer environment.
Advanced algorithms are also playing an increasingly pivotal role. Mathieu Mollin cites the example of tubing and wiring – that is, the routing of cables and circuits within an aircraft’s cabin. As each aircraft is different (cabin configurations, screens, equipment, etc.), the electrical diagram varies from one aircraft to another. “A straight line between two points is rarely a viable option. We have to factor in fire risks, electrical currents and electromagnetic interference. There are numerous business-specific parameters,” he points out. The automation of this type of task, long confined to deterministic algorithms, is becoming increasingly capable of incorporating business constraints.
The two professionals agree on one point: current productivity gains remain limited. “These tools save time on tasks that do not account for the majority of engineers’ time,” summarises Damien Constantin. The real potential for productivity therefore lies elsewhere. But where?
Planning for the agent-based shift today, for tomorrow
Beyond automation, a second wave is emerging with agent-based AI. Here, the challenge is to coordinate complete sequences that can connect different systems (including non-AI systems) and manage end-to-end engineering cycles. Mathieu Mollin outlines the prospect of generative design assistance based on our clients’ rules and methods: “We’re not going to ask an AI to design an entire aeroplane tomorrow. But we can ask it to help us make decisions when cross-system design conflicts arise, or even to predict them in advance so we can avoid them,” he explains.
Damien Constantin adds: “We are already working on this predictive aspect, including taking into account the needs of Manufacturing and Customer Services.”
Far from being an isolated project, agent-based systems are establishing themselves as a transversal enabler that is gradually permeating every component of the system. Future aircraft programmes currently being designed will be developed with these architectures integrated from the outset. Modularity, data access, sharing and agent-based orchestration capabilities: these properties are being defined at the design stage of future PLM systems. The architectural choices we make now will determine the capacity for industrial innovation over the next ten years.
Industrialising AI means taking control of its sovereignty
Our capacity for innovation will depend on our technical capabilities in the field of artificial intelligence. This analogy features regularly in Mathieu Mollin’s discourse: AI is the new electricity, to use the phrase popularised by Andrew Ng¹. And just like electricity at the start of the 20th century, the technology will only realise its potential if production models are overhauled. Replacing a large coal-fired engine with a large electric one did little to transform the industry. The widespread adoption of small electric motors, on the other hand, profoundly reshaped it. The parallel holds true for agent-based AI: its integration requires rethinking processes and preparing teams for new ways of working. Training, change management, redefining approaches: human support remains inseparable from technological deployment.
In this transformation, sovereignty becomes a key concern as soon as AI is adopted for industrial use. Both professionals emphasise this point. National and European sovereignty, as reflected in access to foundation models, is only one aspect of this. The second aspect, which is even more fundamental for manufacturers, concerns the intellectual property rights of the agents that execute business processes. “When an agent takes on tasks at the heart of the business, the company must retain control over it: understanding how it works, knowing how it evolves, and ensuring the continuity of its operation,” states Mathieu Mollin. Control over this intellectual property becomes a strategic asset on a par with patents or technical data.
Damien Constantin expands on this point: “As soon as we move beyond the initial productivity gains and towards a complete overhaul of engineering processes, the architecture for data access and its structuring become crucial. Data sovereignty is therefore being built right now, through the architectural and governance choices made for future PLM systems. Anticipating these choices ensures that manufacturers can continue to carry out their core business, regardless of how the software market evolves.”
Towards the orchestration of specialised agents
Mathieu Mollin anticipates an explosion in the number of hyper-specialised tools over the next three years. As AI reduces the cost of developing software solutions, businesses will be able to equip themselves with a growing number of agents dedicated to specific use cases. The governance of such an application portfolio is becoming a key challenge, one that Sopra Steria is already helping its clients address: standardising the agent lifecycle, managing legacy systems, and redefining the boundary between business functions and information systems.
“We are already orchestrating the operation of these agents, pooling resources and prioritising actions according to the expected return on investment,” summarises Damien Constantin. This dual expertise – in-depth knowledge of business processes on the one hand, and of data and its ontology on the other – forms the foundation of this approach, which is firmly rooted in our clients’ realities.
Sopra Steria is driving this AI transformation first and foremost in-house, across its own tasks and processes. This experience now informs the support it provides to its clients – in both engineering departments and IT departments – as they redefine processes and shape the future of their business lines
¹ A phrase popularised by Andrew Ng, a professor at Stanford, co-founder of Coursera and former Chief Scientist at Baidu, during talks from 2017 onwards.