This is my second article about PLM topics presented at AU 2026. If you missed the first article check this link
Autodesk Agentic PLM: What Exactly Is Being Redefined?
Kevin Schneider’s part of the Fusion strategy session at AU 2026 focused on a practical change in product development. People and an assistant could work concurrently on product information. The BOM could include engineering overrides and content such as adhesives or software that a geometric assembly alone would not describe.
This is meaningful progress. A product definition has to support the information needed to build and maintain a product. Making that definition easier to enrich and work on together is a useful investment.
It also brought me back to a question I have followed for years: how do Autodesk’s different data-management foundations come together? The new Agentic PLM canvas makes this question more consequential because it proposes a common place to investigate problems and initiate lifecycle work.
I see two dimensions that deserve closer examination. One is the relationship among Autodesk’s own data foundations. The other is the connection to the heterogeneous engineering sources that manufacturers already use.
Three foundations behind the experience
Fusion BOM, Fusion Manage, and Vault participate in the strategy with different roles. The distinction is easier to understand when we place them side by side.
| Product | Documented foundation | Question for the connected experience |
|---|---|---|
| Autodesk Fusion BOM | Manufacturing Data Model component and BOM relationships exposed through APS | How does the live engineering definition relate to approved lifecycle records? |
| Autodesk Fusion Manage | Separate cloud PLM product with workspaces, items, BOMs, and lifecycle processes | Which records govern the product change and its release? |
| Autodesk Vault | PDM managing engineering files, items, and versions in a separate server environment | How do cloud workflows retain access to the relevant source state and history? |
The Manufacturing Data Model BOM documentation distinguishes BOM relationships from CAD assembly relationships and allows BOM edits independent of the design. Autodesk also explicitly describes Fusion Manage as a separate PLM product, usable with Fusion or independently.
Autodesk is connecting these environments. The integrated PDM and PLM experiences shown in the sessions is a right move. The most accurate description I can give is distinct data foundations connected through product-specific integrations. A shared user experience brings them closer together, while their relationships still require explicit management.
This continues a question I raised after Autodesk DevCon in 2024: how will the platform connect existing product systems as it develops a more granular data foundation? The agentic strategy increases the importance of understanding that connection.
Vault and the path to Agentic PLM
In my assessment, Vault is a significant strategic constraint on Autodesk’s industry-cloud ambitions. The company must bring engineering information governed in separate server environments into workflows that increasingly depend on cloud access and coordinated action.
Vault represents substantial customer investment in engineering practices and managed information. Those customers need a practical path into the new experience. A successful strategy must make that existing information useful while respecting the environment responsible for it.
The native Vault Connector is a good step. Its documented route pushes items and BOMs from Vault Professional to Fusion Manage. It installs on the Vault Server, requires connectivity and configuration, and treats Vault as authoritative for mapped properties. Changes to those properties in Fusion Manage can be overwritten at the next synchronization.
That establishes a useful integration contract. It also makes the boundary visible. An agent proposing a change in a cloud environment needs to understand which information it can update and which information must be changed through the engineering source.
Another detail is especially relevant to product memory. The documented connector does not synchronize Vault item revision history. It updates the Fusion Manage record to reflect the current Vault version. An investigation requiring an earlier engineering state would therefore need another route to that history.
This describes one supported connector, rather than every possible Autodesk or partner integration. Still, it gives us a specific question for the agentic experience: how does the assistant obtain the history required to explain a change when that history remains in the source system?
Autodesk already describes Assistant capabilities in Vault, as well as on-premises and cloud-hosted Vault deployments. These investments matter. My architectural assessment is that cloud hosting alone does not establish a common data model or resolve the lifecycle relationships between the products.
The strategic issue is how much connection work remains with the customer and how consistently Autodesk can deliver the experience across those deployments. Vault’s long-term place in the industry cloud deserves a clearer explanation.
Where the engineering context comes from
Heterogeneous engineering sources broaden this assessment. Autodesk promotes multi-CAD integration, and Vault has a role in managing mixed engineering information. However, the two sessions gave me less visibility into the complete path from diverse engineering sources to Fusion BOM, Fusion Manage, and the new agentic canvas.
The question is the depth and continuity of the connection. A translated model can support visualization. An extracted structure can support a BOM. A continuously connected engineering source must also preserve the relevant identity, configuration, revision, and approval state as work changes.
The Fusion BOM help describes its scope around items in a Fusion design. For a manufacturer working across several authoring systems, I want to understand how those external engineering definitions enter this environment and remain connected to their sources.
Consider a hypothetical product with an Autodesk Inventor assembly, a supplier’s SolidWorks subassembly, a PCB maintained in another engineering system (eg. Altium Designer exposed in Altium 365), and a software release. A supplier proposes a motor replacement. The team opens a canvas to evaluate the change.
The assistant needs to establish which configurations use the motor and which revision the team is evaluating. It also needs to distinguish an imported representation from information still governed in the supplier’s engineering environment. If the engineer later changes the source assembly, the investigation must remain understandable in relation to that new state.
This would be a revealing demonstration of the strategy. It would show how a shared experience handles the reality of engineering information distributed across tools, companies, and repositories.
What dependable connection would look like
A single physical database is one possible approach to integration. Reliable federation can also connect systems that retain their own records and authority. I would like to assess Autodesk’s direction through the behavior a customer can observe.
Can a reviewer follow the same component across the engineering definition and the lifecycle record? Can the assistant identify the applicable revision? If a source has changed since an investigation began, does the team see that difference before approval?
The workflow also needs an understandable outcome when updates span systems. A completed BOM edit and a pending lifecycle action are different states. The team should know what succeeded and what still needs attention.
The APIs and connected experiences shown at AU create opportunities to solve these problems. The next step is demonstrating the relationships through a complete workflow, including the source update and the governed release.
What is my conclusion?
Autodesk has useful data-management assets and is building a more accessible working experience around them. The strategic undertaking and a major challenge is to make those assets behave dependably together and connect them to the engineering environments customers already have. Especially when all these sources are located in different places and mixed on premise /hosted/cloud environments.
Autodesk Vault makes the deployment and history questions concrete. Heterogeneous sources make the scope larger than Autodesk’s own portfolio. Together, they provide a practical test of the Agentic PLM ambition: can a team carry engineering context through a product decision and back into the systems responsible for the change?
Just my thoughts. YMMV.
Best, Oleg
Disclosure: I am CEO and co-founder of OpenBOM. My work on product data and collaboration informs this analysis.
Note my other articles about AU 2026
From Digital Thread to Product Memory in Operations
AU 2026: MCP Connects the Tools but Who Connects the Product Context
Invisible PLM and the Engineering Controls That Still Matter
AU 2026: The Platform Question Is No Longer APIs. It Is Context.
AU 2026 and the Autodesk Bet on Context for PLM
