AU 2026 was a great week, and I paid for it. Every day ran from 6am to 10pm, some started earlier. Four conference days at 16 hours each is 64 hours, and my body sent me the bill on the way home.
Here’s how the week went.
Sunday: Travel and The Wizard of Oz at the Sphere
Travel day. We landed in Las Vegas and took a Zoox autonomous cab. No driver, and we could connect over Bluetooth and play our own music for the ride. What impressed me was how it handled the busy traffic and the construction zones. Vegas traffic is not an easy test. Fun connection for the Autodesk crowd: former Autodesk CEO Carl Bass became Zoox’s executive chairman in 2018. Then check in, find the badge line, and head to the Sphere for The Wizard of Oz.
Even the lobby blows your mind. The atrium hits you with visuals and sound before you ever find your seat, including a 50-foot-tall holographic display. Sphere also has humanoid robots named Aura that greet guests, but they weren’t out during the Wizard of Oz showings.
It was amazing. They remastered the film and expanded the frame out to fill the Sphere’s full wraparound resolution. Then they added everything else. Haptics in the seats. Wind. Blowing leaves, glitter, and fake snow drifting through the Sphere. Big mechanical butterflies and flying monkeys overhead. When the apple trees shook in the movie, foam apples dropped on the audience.
What they did to the original film is a story on its own. The 1939 movie was shot in a nearly square 4:3 frame, which looks silly stretched across a screen that wraps over and around you. So Google DeepMind, Google Cloud, Sphere Studios, Magnopus, and Warner Bros. Discovery rebuilt it. First, tuned versions of Google’s Veo, Imagen, and Gemini models upscaled the tiny celluloid frames to 16K x 16K. Then AI outpainting extended every scene past the edges of the original frame. The hard part was camera cuts. When the Cowardly Lion jumps out at Dorothy, the Scarecrow, and the Tin Man, the original cuts back and forth between them. At the Sphere, all of them had to be on screen at once, so the AI had to generate the characters the camera never showed. To keep it faithful, the team dug through the archives for the shooting script, production illustrations, photos, set plans, and the score.
The building itself is just as wild. The Sphere is 516 feet wide and 366 feet tall, with about 17,600 seats. Inside, the screen is 160,000 square feet at 16K x 16K, 240 feet tall, and made up of 64,000 LED tiles. Hidden behind it are roughly 1,600 Holoplot speaker modules with 167,000 individually amplified drivers, using beamforming so every section gets its own mix. About 10,000 of the seats have haptics built in. And the outside, the Exosphere, is another 580,000 square feet of programmable LEDs. I spend my days talking about AI taking over the heavy processing so people can focus on judgment. Here was AI doing exactly that on a 1939 classic, with a room full of engineers deciding what looked right.
It’s a 1939 film running inside one of the most technology-heavy venues on the planet, and somehow it works. A great way to start a week that would leave no room for anything else.
Day 0: Special Invite Track and the Product Teams
Monday was the special invite track. I spent the day in product team meetings with the AutoCAD and Civil 3D teams. There was also a Leadership Track running at the same time.
This is the part of AU most attendees never see, and for me it’s the most valuable. You get to sit across the table from the people building the software and tell them what actually happens when 500 users open it every morning. They listen. Not everything lands on the roadmap, but the conversation matters.
The day closed with the Speakers Reception. Always a good room. Lots of people I only see once a year, all comparing notes on whose demo broke and whose didn’t.
No easing into this week.
Day 1: A Full Room, a Crashed Demo, and a Family Reunion
Day 1 started at 5:30am. Not by choice. When you have two hours of material and a 30 minute slot, you get up early and cut. My handout ended up at 40 pages, and it’s in the downloads on the AU 2026 site along with the presentation. I’ll post both here on the blog too.
Breakfast with over 10,000 people is an experience.
The Keynote
Andrew Anagnost and Raji Arasu opened with the main stage keynote on Autodesk AI and Industry Clouds, with live demos showing how AI is absorbing the repetitive work across design, make, and operate. The fireside chat with Clara Shih of the New Work Foundation on agentic AI was a highlight, drawing on her AI work at Salesforce and Meta.
The Autodesk Tandem team had their live demo crash on stage. I felt for them. That’s always the risk with live demos, and I’ve stood in those exact shoes. You demo live because it’s real, and sometimes real bites you in front of thousands of people. Respect for doing it live at all.
My Session: AS1411
My session was at 11:30. I walked in an hour early and a line was already forming, so I asked staff to just let people in rather than have them stand in the hallway. All 280 seats filled, then people started lining the walls. Over 300 in the room, standing room only, and about 100 more turned away at capacity. If you were one of the people left in the hall, I’m sorry. That’s not the experience anyone flies to Las Vegas for.
Apparently CAD management and AI is a hot ticket right now. I believe it.
The short version of AS1411, “The Stack Inversion Is Real: How AI Is Changing What CAD Managers Actually Do”: AI is absorbing the research and processing work at the bottom of the CAD manager’s stack, and the human role moves up into judgment, governance, and ownership. AI handles the evidence. You own the decision.
I walked through how we used AI on Autodesk CER crash data to cut crashes 30% in 90 days across hundreds of users, why 50-70% of crashes in my experience are environment problems rather than bugs, and a 30-60-90 day plan to take home and deploy Monday. Treat AI output like work from a capable intern. Same QA/QC rigor. Plausible is not the same as correct.
The audience was fantastic and the questions were sharp. This was my 34th AU, and presenting to a packed room never gets old.
Part of the overflow problem is how AU handles class selection now. You favorite classes instead of enrolling. People favorite sessions for all kinds of reasons – to bookmark them, to watch the recording later, to keep options open for a time slot. A favorite isn’t a commitment. So speakers have no real way to know who will show up, and AU has no real way to size the room. My session proved it. Enrollment gave you a number you could plan around. Favorites give you a guess.
The Best Moment of the Day
Then something happened that I’ve never seen in 34 AUs. After the session, four people came up and showed me CER solutions they had already built. Same day. Working solutions they planned to deploy as soon as they got back to the office.
And it kept going. At a reception that night, someone showed me the CER analysis solution they built after attending my class, and they’re implementing it Monday. Boom. That’s exactly what I hoped for. Not applause, not survey scores. Someone taking the information and actually building something with it.
Usually the best you get after a class is “great session, I’m going to try that.” This was people walking up with the thing already done. That’s the stack inversion in real life. The gap between hearing an idea and building on it used to be weeks. Now it’s hours.
My braindate on using Autodesk CER crash data to solve crashes had six people, and the discussion was excellent. There’s real hunger out there for turning crash data into answers instead of tickets.
Testing Lidar Between Sessions
I spent the day walking around with a tennis ball sized lidar attached to my iPhone, capturing point clouds between sessions. It’s the AuraGo from Q3D Sensing, 30 meter scan range, about $4k. Comparable lidar units start at $12k and go up from there, so I’m putting it through its paces. More on that later.
The Rest of Day 1
The afternoon included sessions with the Autodesk Agentic AI team and product managers. I have thoughts. I’ll write those up once they’ve had time to settle.
Honestly though, the best part of Day 1 wasn’t on any schedule. It was the sheer number of friends and former colleagues I ran into. The conversations, the hugs. It felt like a family reunion, because after 34 of these, it kind of is.
Day 2: Keynote, Classes, Receptions
Wednesday started with the industry keynote and breakfast with 12,000 people.
Then more classes and more receptions. Four receptions, all at the same time. Somehow.
This is the day the walking starts to catch up with you. The Venetian hallways don’t get shorter.
Day 3: Classes and the AU Party with Rebelution, and.. post party Freemont Street and a Bug
Thursday was the last round of classes, then the AU Party at Brooklyn Bowl with Rebelution playing. Great band, great room, and a crowd of CAD people who had earned a night off.
I stayed later than I should have. No regrets.
Getting Home: 36 Hours in Bed
Sunday through Friday I averaged over 20,000 steps a day. By the time I got home my legs were done.
The rest of me wasn’t far behind. I came home with a really bad cold and was flat on my back for 36 hours. My wife got it too. I guess not everything stays in Vegas.
I’m not surprised. Sixteen-hour days, 20,000-plus steps, recycled conference air, a few thousand handshakes, and airports on both ends. That’s the AU tax. If you’re planning for next year, build a recovery day into your calendar after you land. I didn’t, and I should have.
Would I do it again? Yes. Every year. The conversations at AU are the ones that shape what I work on for the next twelve months.
Cheers, Shaan
