At Autodesk University 2025 (AU2025), Autodesk demonstrated a shift toward embedding agents deeply into its platform strategy, hinting at what Beyond PLM calls “Agent as a Service” (AaaS). This isn’t just marketing; during the Autodesk Platform Services (APS) Leadership Forum, speakers emphasized not only generative tools and APIs but also how agents orchestrated via structure-and-context layers could become foundational for engineering and manufacturing software workflows, says Beyond PLM Blog.
A key piece of this architecture is the Model Context Protocol (MCP). Autodesk showed “before MCP” versus “after MCP” slides to illustrate how agents, when mediated by MCPs, can connect more reliably to applications, data, and tools. Rather than having agents directly tied to apps (a brittle setup), MCPs provide context and structure, enabling agents to perform more useful actions across engineering domains.
Another theme was Autodesk’s AI hierarchy. At the bottom are general-purpose LLMs. Above them are spatial and physical reasoning (3D data), then industry-specific reasoning (AEC, manufacturing), and finally company-specific reasoning based on firms’ internal datasets. The idea is that agents will use these layers to provide richer, more relevant results.
Data model openness and extensibility came up repeatedly. Autodesk is planning to build and manufacture data models that support custom properties, geometry extraction, and cross-tool workflows. There’s also talk of analytics dashboards, clash detection, and coordination tools built on APS.
However, there are many open questions. Will agents become more than demos? Will Autodesk allow third-party or partner extensions, or will this be a closed ecosystem? How will customers pay for agent‐oriented services—by API calls, metrics, or outcomes?
Autodesk appears to be aiming to move from being primarily a tools vendor to a platform operator that delivers intelligent, context-aware agent services. Whether AaaS becomes a useful reality depends on execution, openness, economics, and how real customer workflows change.