Engineering systems of record are not being eliminated—they are being subjected to structural testing.

When Reuters reported in September 2025 that Klarna moved away from Salesforce, citing approximately $2 million in annual savings as part of a shift toward internally developed, AI-enabled tools, the reaction was predictable. The narrative quickly solidified into a familiar headline: AI is replacing SaaS.
For engineering organizations, that interpretation misses the core point.
The real question is not whether AI can replace horizontal workflow platforms. It is whether AI makes it economically and operationally feasible to replace vertical decision-making systems like PLM.
Horizontal systems mainly coordinate standardized workflows across departments. Vertical systems incorporate domain ontologies and oversee critical decisions—such as product structures, configurations, lifecycle states, and compliance—within a specific engineering discipline.
That difference shapes the debate.
The Klarna case is not fundamentally about AI displacing enterprise platforms. It is about what happens when automation interacts with fragmented knowledge structures. Before AI can add value, knowledge needs to be organized. Before ontologies can speed up decision-making, they must be managed.
This is where the engineering view diverges from the headline narrative.
Fragmentation before automation
AI undeniably reduces the cost of developing and modifying software. Code generation accelerates prototyping. Configuration becomes more intuitive for users. Internal tools can be deployed faster than ever before. In environments with standardized workflows, this greatly impacts internal build economics.
But PLM is more than just a workflow engine.
PLM manages product structures, bills of materials, configuration logic, effectivity, change authority, regulatory traceability, and lifecycle accountability throughout years of development. These capabilities are embedded in ontologies and governance logic that mirror established engineering disciplines. They serve as structural commitments to preserving product integrity over time.
The core issue revealed by the Klarna story is fragmentation. Companies accumulate SaaS systems over time: CRM on one platform, analytics on another, documentation stored elsewhere, and operational artifacts spread across various tools. Each system has its own schema and terminology. Integration layers grow more complex. Semantic consistency diminishes.
AI does not fix fragmented ontologies. It amplifies their inconsistencies. Feeding duplicate or conflicting data into an intelligent system does not clarify anything. It leads to plausible ambiguity.
For engineering organizations, this is not just hypothetical. Automating on weak data foundations raises systemic risk. Knowledge needs to be unified before accelerating.
The cost of coding has gone down. The cost of stewardship has not.
Infrastructure is not ontology
Much of the public commentary groups enterprise software into one category. This conceals a key architectural difference.
Horizontal platforms offer infrastructure features such as workflow engines, object frameworks, API-enabled digital threads, and access controls. Vertical platforms include domain ontologies: configuration rules, lifecycle states, compliance constraints, and change governance.
Some modern PLM solutions are built on horizontal foundations. For example, Propel runs on Salesforce. This architecture highlights the difference: Salesforce provides infrastructure services, while Propel manages the product ontology and lifecycle governance.
Infrastructure can be redesigned.
Domain ontologies—such as configuration logic, effectivity rules, and lifecycle discipline—cannot be casually recreated without accepting long-term responsibility for their accuracy and development.
Replacing a horizontal workflow layer requires integration effort. Replacing a vertical governance layer involves a structural transfer of accountability.
That distinction is about economics, not just words.
Total cost of ownership is structural
The appeal of AI-powered internal builds is their immediate efficiency. However, the economics of engineering systems work over longer periods.
When organizations license mature COTS engineering platforms, they transfer responsibility for roadmap updates, regulatory compliance, security, interoperability, and backward compatibility across product generations to the vendor. These responsibilities are incorporated into the vendor’s operating model.
An internal build takes responsibility for all these aspects.
AI reduces the effort needed to start development but does not lessen the effort required to maintain architectural coherence over fifteen years of product growth. It does not eliminate the complexity of integrating systems like ERP, MES, ALM, quality systems, and supplier networks. Nor does it prevent institutional fragility when key domain architects leave.
The relevant comparison is not licensing fees versus development sprints. It is license fees versus ongoing governance, integration resilience, and compliance accountability.
Engineering leaders recognize which factor is more difficult to control.
Governance before distribution
PLM decisions go beyond PLM itself. They impact manufacturing execution, supply chain planning, quality systems, service documentation, and regulatory reporting. This interconnected flow of product information is commonly called the digital thread.
Vertical depth without horizontal sharing creates silos. Horizontal integration without vertical discipline results in semantic drift.
AI accelerates the distribution process, which also increases the cost of weak governance. Faster ambiguity is more dangerous than slower ambiguity. In an AI-driven environment, structured and opinionated record-keeping systems become even more essential.
A legitimate strategic objection
It would be too simplistic to dismiss the counterargument. AI greatly lowers the barriers to developing internal systems. Large companies with concentrated engineering talent might reasonably choose to develop certain capabilities in-house.
However, the challenge of PLM and other engineering systems of record is not their initial construction. It is maintaining coherence over time.
These systems evolve with product portfolios, regulatory landscapes, acquisitions, supplier changes, and shifts in manufacturing strategy. They must ensure traceability across decades. AI may speed up adaptation. But it does not assume regulatory liability, guarantee ontological integrity, or absorb operational risk.
Those responsibilities remain organizational.
The key question isn’t whether AI can help build parts of a PLM-like system. It can, or it probably will someday. The real question is whether the enterprise is ready to take full lifecycle accountability forever.
For most engineering organizations, that remains the main constraint.
What the Klarna case actually means
The Klarna example does not mark the end of SaaS, nor does it invalidate the structural reasoning behind vertical engineering platforms. It highlights a dislike for fragmented knowledge structures in an AI-driven environment.
The broader market environment amplifies this tension. In February 2026, Dassault Systèmes’ shares fell sharply—over 20% drop in a single session—after the company posted fourth-quarter revenue growth of just 1% year over year and guided 2026 growth below expectations. This is a €6+ billion industrial software company with strong margins and leading positions in CATIA, SOLIDWORKS, and the 3DEXPERIENCE platform. Despite a profile that suggests stability, markets clearly reacted as if the SaaS narrative were broken.
The reaction was less about solvency and more about growth expectations in an AI-driven capital environment. As investors shift toward AI infrastructure and hardware exposure, even highly profitable industrial software companies are being tested to see whether growth is incremental rather than exponential. The implicit question becomes: if AI can increasingly generate, simulate, and optimize designs, what is the defensibility of traditional high-margin software layers?
This does not mean industrial software is obsolete. It indicates that the market is re-evaluating whether these platforms are infrastructure utilities, workflow layers, or deep ontological systems of record. Vendors that merely coordinate execution may face compression. Vendors that control product meaning and embed lifecycle discipline maintain structural relevance—provided they articulate that value clearly.
For engineering leaders, the architectural question is no longer simply build versus buy. It is where product truth is governed, how it is communicated, and who bears responsibility when governance fails.
AI reduces the friction of building software. It does not eliminate the obligation to preserve meaning. And in engineering, meaning—embedded in configuration logic, lifecycle state, and change authority—remains the core of value.