Autodesk Reveals Cross-Cloud AI Orchestrator to Unify Its Three Industry Platforms

At Autodesk University 2026 in Las Vegas, Autodesk on Tuesday previewed what it calls the next generation of Autodesk Assistant — a cross-product agentic AI layer that spans all three of its industry clouds simultaneously and routes tasks to different AI models depending on what each job requires. The announcement marks a structural shift for the company's AI strategy: from per-tool assistants embedded inside individual products toward a unified intelligence layer that holds project context across teams, products, and disciplines.

Autodesk Assistant is already generally available today inside Fusion, Forma, and Flow — the three clouds covering design and manufacturing, architecture and construction, and media and entertainment, respectively. What was previewed at AU 2026 is a substantially different architecture sitting above those products: one that understands not just the question a professional is asking, but the project and constraints behind it.

The AI Orchestrator: Dynamic Model Routing at the Core of the Architecture

The most technically significant component of the announcement is a layer Autodesk calls the AI Orchestrator. According to Autodesk's announcement, it evaluates available AI models on the fly based on a task's specific requirements — accuracy, speed, data security, and cost — and selects the appropriate model for each step rather than feeding all work through a single foundation model.

The implication is architectural. Autodesk is not betting its enterprise AI strategy on building or owning a proprietary large language model. Instead, it is building a model-agnostic routing layer — an abstraction that can draw on different AI providers or specialized internal models depending on what the engineering task demands. This positions Autodesk closer to what Siemens calls an "industrial AI orchestration" platform than to a vendor shipping a single AI model. The competitive advantage Autodesk is describing is not superior model capability; it is decades of proprietary domain data — geometry, physics constraints, fabrication history, project schedules — that general-purpose AI cannot access and cannot replicate by reasoning alone.

Andrew Anagnost, Autodesk's president and CEO, framed the argument during the event: "The constraint in our industries has never been about ideas. It's capacity. And for Design and Make professionals, 'probably right' is wrong because their work has consequences in the real world."

This grounding matters in ways that are particular to physical industries. A hallucinated answer from a chatbot is correctable. A hallucinated structural calculation, a wrong machine-tool path, or an incorrect BIM clash detection result can propagate into fabrication drawings, procurement orders, and construction site decisions. Autodesk's argument is that its combination of project data and task-appropriate model selection reduces that risk in a way that a horizontal AI assistant cannot.

Personal Pulse, Assistant Spaces, and Assistant Builder: The Three Pillars of the Next-Gen Experience

Built on top of the AI Orchestrator, Autodesk has structured the next-generation experience around three operational priorities.

The first is understanding what needs attention. A feature called Personal Pulse surfaces prioritized issues and insights from across a professional's active projects — surfacing schedule slippage, specification conflicts, or downstream fabrication impacts before they escalate. Alongside it, Assistant Spaces creates a focused collaborative environment where teams and their agents can investigate a problem together, see the downstream effects of a change, and coordinate action across disciplines without switching between separate tools.

The second priority is acting across workflows. Rather than describing a problem, the next-generation Assistant is designed to bring forward the precise Autodesk capabilities needed to address it — connecting to simulation engines, manufacturing prep tools, or production pipelines through natural language or voice, and producing reviewable outcomes rather than recommendations that a human must then manually execute.

The third — and likely the most significant for enterprise adoption — is what Autodesk calls Assistant Builder. This allows organizations to connect their own AI agents, internal tools, and partner services to Autodesk's project context layer. Support for third-party agents through the Design & Make Marketplace is planned as part of the 2027 rollout.

The practical consequence for enterprise customers is a platform that can incorporate domain-specific agents — a structural analysis agent, a procurement pricing agent, a sustainability compliance agent — alongside Autodesk's own capabilities, all sharing the same project data context.

Domain-Grounded AI Versus Horizontal Competitors

The enterprise agentic AI space is congesting fast in September 2026. Salesforce unveiled seven named purpose-built AI agents on September 11, covering customer service, sales, HR, commerce, and supply chain, layered over a Trusted Enterprise AI Harness governance framework that includes a cross-platform AI Control Plane. Salesforce reported 7 billion Agentic Work Units delivered across Agentforce and Slack, with 3.2 billion in Q2 2026 alone — all figures the company itself reported.

Siemens announced Intelligence Center X in June 2026 — industrial AI orchestration software designed specifically to connect enterprise manufacturing data, workflows, and AI agents across its Xcelerator portfolio. Siemens cited a 95% reduction in manual effort at one early customer and 85% faster production issue resolution at another, though both figures come from the company's own pilot reporting.

Autodesk's differentiation from both is not the presence of an orchestration layer — that is now table stakes — but the scope and specificity of the domain context each platform brings. Salesforce operates across business processes and customer-facing workflows where the cost of AI error is typically a bad customer interaction. Siemens focuses on manufacturing execution and PLM, with strength in factory-floor and EDA workflows. Autodesk is the only platform spanning the entire design-to-make lifecycle for three distinct industry verticals simultaneously — architecture and construction, product design and manufacturing, and media and entertainment production. The Forma cloud, for instance, is designed as the successor trajectory for Revit users — connecting the BIM data model that is foundational to large AEC projects to an AI layer that can reason over that data at scale.

Read more: Why verification, not capability, is the real bottleneck for agentic AI deployment

What Is Not Resolved and What Remains a Preview

The next-generation Assistant features announced at AU 2026 are not generally available. Autodesk confirmed that rollout timing, subscription terms, and regional availability will be announced in 2027. The standalone agent experience that Autodesk Senior VP of Research Mike Haley described as "an agent-first experience" for workflows running between and around Autodesk products is also planned for 2027.

Several product-level features previewed at AU are similarly future-dated. AutoConstrain and AutoTimeline — AI-driven tools that convert imported geometry into editable parametric parts in Fusion — were previewed as planned rather than shipping. AutoAssemble, which partially automates the assembly workflow, follows the same trajectory. Autodesk's description of Inventor as the first Fusion Connected Client, bringing that historically desktop-only CAD application into the cloud-connected product lifecycle, is a meaningful architectural commitment, but its rollout schedule was not specified.

There are also genuinely open questions about how these systems perform at scale. Autodesk publishes AI Transparency Cards for its assistant features, disclosing data and technologies used, safeguards applied, and known limitations — a transparency practice that reflects the high stakes of AI errors in physical-world engineering contexts. But no independent benchmarks for the agentic features exist yet, because the features are not publicly deployed. The customer evidence cited at AU 2026 — Norconsult's assessment of cross-product context value and the Haas Factory Team's use of Fusion for NASCAR part manufacturing — speaks to the existing product, not the next-generation agentic system.

In Autodesk's own 2027 State of Design & Make Report, the company says 70% of Design and Make leaders expect AI to reduce repetitive work, giving professionals more capacity for judgment and expertise. That survey is Autodesk's own research, not an independent finding — but the directional signal aligns with what is driving the architectural investment being previewed this week.

What Comes Next

The industry pressure Autodesk is responding to is real and measurable. Enterprise software companies that cannot show a credible cross-product AI architecture by 2027 risk losing the attention of customers who are standardizing their AI infrastructure decisions now. The AU 2026 preview positions Autodesk alongside Siemens and Salesforce as one of the legacy enterprise software vendors making the bet that domain depth — not LLM access — is the defensible basis for enterprise AI value. Whether the AI Orchestrator's multi-model routing delivers meaningfully better engineering outcomes than a well-prompted general-purpose model in practice, and whether independent evaluations of the GA system in 2027 confirm the architecture's claims, will determine how credible that bet turns out to be.