By: Nate Bek
Vegas felt like an odd place for a conference about intelligence. The artificial kind, or otherwise. But maybe that made it the right setting.
The slot machines never stopped blinking. The escalators kept funneling us between velvet-curtained ballrooms and DJs and mimosa towers to the floor of a city built to overwhelm. There’s something about putting thousands of people in a place designed to keep you distracted — and asking them to think about the future.
Human[X] was the first AI conference of its kind, but it didn’t really feel like a debut. It felt more like a checkpoint. I came in with a few themes top of mind: robotics, AI labs turning into real institutions, and how venture is reacting to the next wave. Those threads ended up tying most of the sessions I attended together.
From left: Hugging Face, Co-founder; Rachel Metz, AI Reporter at Bloomberg; Arthur Mensch, Co-founder & CEO at Mistral AI; and Metz.
The most grounded — and maybe most striking — observation came from Thomas Wolf of Hugging Face.
"I'm pretty sure 2025 is going to be the year where robots start to work."
Not in theory. In kitchens. In warehouses. On wheels. Reinforcement learning breakthroughs are beginning to land outside the lab, and there’s a new confidence about it. He also flagged what’s missing: openness.
“They're all closed-source… In the future, where robots are everywhere, I would love a part of them to be open so that we know exactly what's inside. Maybe we even build them ourselves."
That tension came up in other corners, too. Brad Porter, founder of Collaborative Robotics, was clear about the practical challenges of humanoid robots. Drawing on his time at Amazon:
“We needed robots that could work collaboratively in and around humans. It was not clear how you were going to safety rate these 300-pound mechanical things that can fall over really easily."
What stood out to me was how rarely this kind of specificity comes up in general AI discourse. These weren’t hypotheticals. They were design constraints from people who’ve already tried and course-corrected.
Alfred Lin of Sequoia framed it well — almost offhandedly:
“The desert is not designed for humans. Just like highways and space.”
Form should follow function. The future of robotics may feel novel, but its success will come down to how it adapts to spaces we already occupy.
From left: Mike Krieger, CPO at Anthropic; Moderator: Alex Heath, Deputy Editor at The Verge; and Kevin Weil, CPO at OpenAI.
Mistral’s session with Arthur Mensch made this shift clear. He spoke about deploying vision models on the edge.
“When the robots [are] in the outer world, it won’t have a very strong connection… you want to have low-latency decision making.”
It’s a reminder that not every AI breakthrough lives in the cloud. Edge environments are where robotics will have to operate, and some of the more forward-looking labs are already planning around that.
He also spoke about Mistral’s recent partnership with Helsing, and the demand for sovereign infrastructure in Europe:
“There's a lot of demand to have a regional cloud because that brings more sovereignty and leverage from an economic perspective… As it turns out, we're pretty good at operating GPUs.”
Labs are becoming more than R&D centers. They’re shaping policy, forming defense partnerships, and influencing national strategies.
Anthropic’s Mike Krieger talked through another facet of this transformation — the delicate balance between building products and staying aligned with partners:
“I called basically all of our leading coding customers to give them a heads up that we're launching Claude Code… We're hearing from people using both.”
He was transparent. It was a small window into what it takes to navigate product expansion when your customers are also your ecosystem.
There was a lot of quiet confidence in the VC session with Navin Chaddha, Lauren Kolodny, and Martin Mignot.
“We're still not at the peak of the hype cycle,” said Chaddha. “VCs have enough money.”
He cited $307 billion in dry powder. Valuations, in his view, still have room to move. LPs remain focused on DPI. “It’s a financial services business,” he said toward the end — a reminder that beneath the storytelling, this is all still portfolio math.
Martin Mignot pointed to the upside:
“We can have trillion dollar VC-backed companies when they go public… A lot of value is being captured by VCs.”
The liquidity window may be narrow, but the bets are only getting larger. And many of them are moving into real-world infrastructure — robotics, autonomy, defense.
In between sessions, I kept thinking about how far we’ve moved from the “chatbot moment.” Human[X] was filled with serious people talking about serious problems. Model architectures were mentioned less than operational realities.
Vegas was a strange backdrop. Too loud. Too bright. Too artificial. But also kind of perfect. A place built on simulation, where beneath the surface you start to feel what’s real.
