Road to AU 2026: what I am presenting in Las Vegas

This is not a normal week. On Tuesday, Autodesk University 2026 kicks off in Las Vegas, and on Thursday the 17th at 4:30 PM I take the stage with session MFG3731, Vault on Autopilot: AI-Driven PDM Automation Across Lifecycle, Revisions, and ERP. A 60 minute Technical Deep Dive in the Design & Manufacturing track.

I have spent months not telling the full story, so here is the honest summary of what will happen in that room and why I believe it is worth your hour if you are at the event.

What I am presenting

The session tells how we built VaultPilot together with Avant Leap: an AI agent that lives inside Vault Professional and runs real PDM operations from natural language. Search, lifecycles, properties, bulk operations, BOM, ECOs, jobs, ERP integration and reports. Not a chatbot that answers questions about your Vault: an agent that works in it.

And precisely because it works in it, half of the session is not about what the agent can do, but about what we do not allow it to do. Dry-run with confirmation before every write, validation against the live Vault schema, verification of every operation with read-back, hard limits on bulk operations and a full audit trail. The thesis is simple: autonomous does not mean no human. The agent runs the workflow; you authorize the write.

Why it ships with no hype

AI talks usually show the demo that went well. I am going to show the development log: the raw numbers of two months of work against a Vault Professional 2026 instance, including the writes our gates stopped and the times the API reported success and the read-back proved otherwise. It is not a production success rate; it is the evidence of why control matters more than magic.

If you manage an engineering team that lives in Vault, or you are an Autodesk partner who gets asked about AI every week, that hour will give you the criteria to separate what already works from what is still a promise.

What you take from the hour

I am not going out there to sell a tool; I am going out there to hand over a method. The session includes the formula we use to prioritize what to automate first: frequency times manual minutes times people, divided by effort and risk, with risk scored 1 if the operation is reversible and 5 if it is not. It is the same math that separates an automation that pays for itself in weeks from a project that should never have started.

We also talk about the starting point almost nobody measures: in the field we have observed teams where each engineer spends between 3 and 8 hours a week on repetitive PDM operations. That is not an instrumented figure, it is field observation, and that is exactly why in the session I insist on measuring yours before automating anything.

Where to find me

I will be in Las Vegas for all three days of the event. If you want to talk about agents, Vault, automation without AI, or a process that is stealing hours from your team, message me on LinkedIn or find me at the event braindates. The best conversations at AU never happen on stage.