At most enterprise software conferences this year, you hear the word AI so often it starts to become wallpaper. That was not quite my experience going through the ACE 2026 keynote materials from Aras. What struck me instead was a different kind of argument: Aras talking about where it came from as a way to explain where it is going. If that continuity holds up under scrutiny, it is a more interesting story than any product announcement.
For those who is asking what is Aras, here is a quick summary I captured from the presentation of Aras’ CMO Josh Epstein during Threaded event at ACE 2026
I have been writing about this space for a long time, and one question has stayed with me through the last year of AI-in-PLM conversation: are vendors actually evolving their architectures to support AI in a meaningful way, or are they building co-pilots on top of systems that were never designed for the kind of intelligence they are now promising? That question is what made me pay close attention to the three keynotes from Leon Lauritsen (Aras CEO), Igal Kaptsan (Aras CPO / SVP of Product and Strategy), and Rob McAveney (Aras CTO).
My takeaway is that Aras is not trying to reinvent itself from scratch for the AI era. It appears to be extending something that was always part of its identity. And that is either a genuinely compelling long-term strategy, or a familiar story being told in new language. I think it is more the former, with one significant architectural caveat I will come back to.
Leon Lauritsen’s Keynote: Complexity Is the Market
Leon’s opening message was titled “Our Current Reality and Moving Forward” and it established the business problem Aras is positioning itself to solve. The frame is not just AI adoption. It is something bigger and harder: expanding product and regulatory complexity across the full lifecycle of modern products.
The slides walked through what that complexity actually looks like in practice today. Regulatory complexity is no longer a compliance checkbox; the ability to manage it is becoming a competitive advantage. Products now span mechanical, electrical, software, systems engineering, simulation, quality, compliance, manufacturing, supply chain, and service domains.
Leon’s slides literally built this picture layer by layer, showing how classic PDM and PLM covered a narrow slice of that reality: CAD, BOM, documents, and a basic as-built record. The rest of the product lifecycle was either scattered across disconnected tools or not captured at all.
What I found important in Leon’s framing is that he was not selling simplification. He was saying the world is getting more complex, and the answer is a platform that can manage that complexity with structure, extensibility, security, and governance at scale. The slides built toward a key question: “What do we need?” and then progressively added requirements: extensible data model, workflow and business logic management, security and governance at scale, integration and federation, performance at scale, reimagined user experience, AI governance, and finally agentic intelligence.
That progression matters because it shows where Aras thinks the gaps are in the market. Not just AI features. An entire stack that can hold together as complexity grows.
Igal Kaptsan’s Keynote: Adaptive PLM and the Dependency Graph
Igal’s keynote was the most explicit product direction statement of the three, and the phrase at the center of it was Adaptive PLM. The definition he offered on the slides is worth capturing precisely: Adaptive PLM evolves with your data, processes, and business. It builds on existing data, learns and adapts as changes happen, connects data into a live network, and uses AI to support decisions and what-if analysis.
That is not a minor update to how PLM has traditionally been framed. Most PLM implementations are defined at deployment and then maintained. The idea of a system that continuously adapts with the data and the business is a different architectural posture entirely.
The cloud and platform strategy Igal described is also relevant here. Aras Cloud Innovator is being positioned as a platform as a service for digital thread and PLM solutions. The keynote walked through public cloud SaaS, customer-managed deployment, containerized architecture (coming second half of 2026), Gov Cloud, DevOps, and the CIAM identity layer. Innovator Edge was highlighted as the mechanism for extending the digital thread beyond the core PLM environment into the broader digital ecosystem, with Edge AI coming in Q3.
But the section that caught my attention most was the dependency graph sequence. Igal dedicated a substantial portion of the keynote to this, and the language was direct: “The future of PLM is tied to our ability to enrich our digital thread with the full context of dependency relationships across the full spectrum of information.” A slide titled “What You Don’t See Is Where Risk Lives” made the business case sharply: you make a change, you validate what you can see, everything looks connected, but impact unfolds beyond visibility in other systems, teams, or domains. The hidden cost of that limited visibility, the slides argued, is one of the largest unaddressed costs in manufacturing.
The proposed answer was a dependency graph layer that enriches from basic connections to rich dependency relationships, and an AI-driven paradigm Igal called Adaptive PLM in action: a system that autonomously surfaces all impacts, saving engineers time by triggering semantic expansion across the graph, predicting cost and schedule changes, and generating new relationship edges from uploaded specifications.
That is a significant architectural ambition. And it leads directly to the most important question in the strategy, which I will come back to.
Rob McAveney’s Keynote: The Agentification of PLM
Rob’s keynote carried the title “The Agentification of PLM: Rethinking how people work together to make things in an age of HAX and AAX.” HAX and AAX refer to human-agent and agent-agent interaction models, and Rob used them to lay out a fundamental shift in how PLM user experience will be redesigned.
The interaction model breakdown was one of the clearest parts of the keynote. Rob described a spectrum moving from conversational, question-and-answer interactions, through task-driven focused workflows, to multi-step processes governed by rules, to collaborative shared workspaces, to fully autonomous monitoring. The discovery layer was similarly structured: insights on demand through natural language analytics, proactive discovery through ThreadRAG and prompt-to-visualization, and invisible discovery through zero-UI triage and monitoring agents.
The agent-centered design message was direct: AI agents are the new user. Traditional interfaces minimized. Natural inputs and context take over. Hyper-personalized, generative, adaptive UX experiences become standard. Transparency is mandatory.
The governance slide was one I want to highlight specifically because it is easy to skip past governance language as obligatory enterprise boilerplate. Rob’s slide was more pointed than that: “Clear operating rules, traceable decisions, and controlled authority.” The roadmap extended this into a Horizon 3 labeled “Trusted Autonomy” projected to 2028, explicitly framed as enabling trusted autonomy across the full product lifecycle. The word trusted is doing a lot of work there, and I think it reflects a genuine architectural commitment rather than just a marketing qualifier. Aras appears to be saying that agents without governance are not something it will ship as a finished product vision.
Knowledge graphs also appeared explicitly in the Rob keynote, connecting directly to Igal’s dependency graph work. The arc of the three keynotes resolves into a coherent picture: dependency graphs and knowledge graphs are the structural layer that makes both adaptive PLM and trusted agentic operations possible.
From Flexible Data Model to Organizational and Product Memory
The part of the ACE 2026 keynotes that I keep returning to is not any individual slide. It is the continuity line between Aras’ original identity and the direction it is now describing.
Aras was widely known in its early days, under Peter Schroer’s original vision for Aras Innovator, for the flexibility of its data model. At a time when most PLM systems were rigid, expensive to reshape, and slow to evolve when a customer’s products or processes did not fit the predefined structure, Aras stood out because it offered a more adaptable foundation. The flexible data model was not just a technical differentiator. It was a philosophical one. PLM should fit the business, not the other way around. Aras also built a strong reputation around seamless upgradability, which was another form of that same flexibility applied to the system itself.
Looking at the ACE 2026 direction now, that original idea seems newly and powerfully relevant. If the future of PLM is moving toward dependency networks, knowledge graphs, context graphs, and AI agents that must reason across lifecycle information, then the challenge is no longer only about storing structured records or orchestrating workflows. It becomes a question of how to represent connected meaning: across products, across disciplines, across organizational memory.
The evolution path looks like this. Flexible data modeling becomes the basis for modeling relationships. Relationship modeling becomes the basis for dependency networks and knowledge graphs. Knowledge graphs become the substrate for AI agents that can navigate, reason, and act across connected product information. At that point, the system is no longer just a database with process management on top. It is moving toward something that preserves not only what changed, but how things are connected, why decisions were made, and how knowledge accumulates around products over time. That is product and organizational memory. And that is the most important strategic concept in the whole ACE 2026 story, even if it was not named that way from the stage.
Now for the caveat, and I think it is an important one.
The dependency graph and knowledge graph vision is compelling precisely because it goes beyond what relational databases were designed to do. Traditional relational architecture excels at structured records, transactional consistency, and well-defined relationships between known entity types. But the dependency graph vision Igal described is fundamentally different. It requires traversing arbitrary relationship paths across large and evolving data networks, performing graph analytics to surface hidden impact, generating new relationship edges dynamically, and doing all of this at the scale and speed that agentic workflows require. That is not a problem you solve by adding a graph layer to a relational core. It requires rethinking how data is stored, queried, and traversed at the infrastructure level.
Aras Innovator today is still fundamentally a relational database architecture. The flexible data model was always relational flexibility, not graph-native flexibility. The move from a richly extensible relational schema to a truly graph-native data layer is not a configuration change. It is a significant architectural investment, and it involves questions that the keynotes did not directly address: What graph database infrastructure is Aras building on or partnering with? How will graph queries interact with the existing relational data model? What is the migration path for existing customers whose data lives in relational tables?
None of this means the strategy is wrong. It may mean the timeline is longer than the roadmap slides suggest, or that the dependency graph capability begins as a layer on top of the relational foundation before a deeper architectural shift becomes possible. That is a legitimate path. But it is worth being clear-eyed about: the gap between a flexible relational data model and a native graph data management system is real, and closing it is one of the harder engineering problems Aras would need to solve to fully deliver on the vision Rob and Igal described.
This is not a reason to dismiss the strategy. It is a reason to watch how Aras executes it, and to hold the vision accountable to the architectural reality as the next product releases arrive.
What is my conclusion?
How Aras Can Stay Different. Walking away from the ACE 2026 keynotes, the question I keep coming back to is not whether Aras is embracing AI. Of course it is. Every enterprise software company is. The more interesting question is whether Aras can connect its historical architectural identity to a future that genuinely requires something different from what PLM systems have traditionally been.
The argument I see in the three keynotes, taken together, is that Aras believes its long-standing strength in flexible data modeling is not a legacy artifact. It is the early expression of an architectural philosophy that becomes more important, not less, as PLM moves toward dependency networks, knowledge graphs, context-aware systems, and agent-driven operations. If the platform that can adapt its data model can also adapt its relational structures toward graph relationships, and eventually toward genuine graph-native data management, then Aras has a coherent story from its roots to its future.
The flexible data model that differentiated Aras twenty years ago may now be evolving into something larger: the ability to model relationships, dependencies, context, and memory at a scale that traditional PLM systems were never designed to reach. If that transition succeeds, Aras is not just adding AI to PLM. It is rebuilding PLM as an adaptive, governed, agent-ready platform capable of capturing and preserving product and organizational memory.
The PLM vendors that remain structurally relevant in the next decade may not be the ones with the most ambitious AI language in their keynotes. They may be the ones whose architecture can actually evolve from systems of records and workflows into systems of relationships, context, and memory. That is a much harder transition. It is also a much more meaningful one. Aras seems to understand this, and ACE 2026 was a visible statement of that intent.
Whether the execution matches the ambition is the question that the next two years of product releases will answer.
Just my thoughts…
Best, Oleg
Disclaimer: I’m the co-founder and CEO of OpenBOM, a collaborative digital thread platform that helps engineering and manufacturing teams work with connected, structured product data, increasingly augmented by AI-powered automation and insights.
