The prostep ivip Symposium has long been a barometer for where engineering and product lifecycle disciplines are headed. This year, the signal was unmistakable. AI has cleared the conceptual stage. The conversations across sessions, roundtables and corridor exchanges were grounded in deployment realities, adoption barriers and the hard work of making intelligent systems earn their place inside engineering workflows.
For practitioners in PLM, digital thread and engineering operations, this marks a meaningful inflection point one worth examining carefully.
AI is no longer a presentation topic, it is a present reality
The most consistent theme across the Symposium was that AI has left the realm of showcases and entered active use. Organizations are running AI-assisted processes in requirements validation, change impact analysis, configuration management and supplier data governance. The language has shifted from potential to performance.
This is significant for engineering leaders to register. The window for passive observation is narrowing. Peers across automotive, aerospace, industrial machinery and MedTech industries are moving through the pilot-to-production transition. Competitive differentiation is beginning to accrue to organizations that have committed to AI-enabled engineering processes not those still running proofs of concept.
The Symposium also surfaced a more nuanced perspective on what this adoption looks like in practice. It is rarely dramatic. Most deployments are augmentative AI working alongside engineers on tasks that are high in cognitive load but low in ambiguity, such as document classification, cross-BOM consistency checks and semantic search across product knowledge bases. The value compounds quietly before it becomes visible.
The case against general-purpose AI in engineering contexts
One of the sharper discussions at the Symposium centered on model selection. The dominant question was whether large language models built on broad, generalist training data are the right foundation for engineering-specific AI applications. The emerging answer, supported by practitioners already deploying AI in PLM and R&D environments, is that they frequently are not.
Engineering knowledge is precise, contextual and organizationally specific. A requirement in a MedTech Bill of Materials (BOM), for example, carries regulatory implications that a generalist model will not recognize. Similarly, a change order in an aerospace configuration management system triggers downstream impact logic that general-purpose reasoning can’t reliably trace. The precision required in these domains demands a different kind of model architecture.
This is where domain-specific small language models (SLMs) are drawing serious attention. Trained on curated engineering corpora and fine-tuned against organization-specific product data, SLMs offer two advantages that matter operationally: they are accurate in ways that Large language models (LLMs) are not, and they are significantly cheaper to run at scale.
SLMs: The economic and strategic logic for engineering teams
The cost argument for SLMs in engineering environments is straightforward. Inference costs for large models queried at the frequency that PLM workflows require are non-trivial. Engineering teams running AI against product data, change histories and compliance documentation are dealing with high query volumes and long context requirements. SLMs, operating at a fraction of the compute cost, make this financially viable at enterprise scale.
The strategic argument is equally compelling. Most organizations of scale have accumulated decades of engineering knowledge: in Computer Aided Design (CAD) libraries, legacy BOM structures, design rationale documents, simulation outputs and supplier records. This institutional knowledge is currently locked away, inaccessible to the speed of modern engineering workflows. SLMs trained on this data don’t create new knowledge. They unlock what already exists.
That reframes the AI investment conversation for CTOs and engineering leaders. The question stops being about how much to spend on frontier model access and shifts to how to structure existing intellectual property as a training asset. Organizations with mature digital thread strategies and well-governed PLM environments have a meaningful head start here.
The strategic takeaway for engineering leaders
The prostep ivip Symposium offered a useful calibration for where engineering AI genuinely stands; past the hype cycle, but still well short of widespread operational maturity. The organizations pulling ahead are those treating AI adoption as a capability-building program rather than a series of isolated technology decisions.
For PLM and R&D leaders, the priority areas that emerged from this year's discussions are practical. Audit existing engineering data assets for SLM training readiness. Identify the two or three workflow categories, such as change impact analysis, compliance traceability and knowledge retrieval, where domain-specific AI will deliver the clearest productivity return. Build the governance scaffolding before scaling.
The Symposium made one thing clear above all else: the leading engineering organizations are not asking whether AI belongs in their workflows. They are focused on getting the model architecture right, the data foundations in order and the organizational readiness built. That is where the real work, and the real competitive distance is being made.
