all I talk about

I've long been interested in automating reality capture with robotics as the prerequisite mapping step required for robots to understand the world with enough precision and accuracy to effectively manipulate it. A lot has happened since I stepped away from construction robotics to spend a few years designing behavior authoring UX for generally capable humanoids. This includes the emergence of VLMs capable of semantic and inherent spatial reasoning for robot action, and world models that act as virtual gyms to evaluate, refine, and collect data for VLA policies. This week I attended Reality Capture Network's R-CON to see if everyone's as Bitter Lesson-pilled as I am, and if they're as curious as I am about agents that understand space. This capability fundamentally alters the reality capture landscape by shifting the focus from collecting measurement data to, quite literally, having conversations with your jobsite. My takeaways: (1) There's tension between licensed surveyors and their methods of control, and the democratization of reality capture services and solutions afforded by newer imaging technologies. The debates were informed and nuanced, and I emerged with an appreciation for both sides. Like most things, the future is not mutually exclusive and will lie somewhere in the middle. There's a more extreme version of this argument where next-gen VLMs will replace all LiDAR-based surveying, but the argument that computer vision will subsume all surveying work is nearly as old as computing itself. (2) There wasn't a compelling argument against the generalist frontier models becoming more capable than models trained on proprietary construction customer data in the areas of safety monitoring, work quantification/progress monitoring, and QA/QC/deviation detection. It was suggested that post-training might always be required, or that the neolabs might outperform the frontier labs in specialized spatial understanding, but I wouldn't bet against the latter if GPT Astra and Gemini Argon are just the beginning of language models with advanced spatial comprehension. (3) Robotic reality capture is the assumed future. I no longer have to waste my breath defending the concept, and can focus on making it reliable enough for obvious ROI. Also: "dog" has become the fully accepted shorthand for reality capture quadrupeds, just as "drone" is shorthand for aerial capture, which I love. (4) The agentic revolution has breathed new life into longtime construction AI players like @openspaceai (2017), @doxel_ai (2016), and @AliceTech_AI (2015). This isn't to suggest they were struggling in the interim, but rather that there's more reason to be excited about these companies now than when they first arrived on the scene. (And imagine the "frontier-native" startups coming next...)

Clog WeWork is finally here clog-online.com/products/clo… with contributions from Kashuo Bennett, Darrick Borowski, Jeremiah Britton, Christian Callaghan, Annie Cosgrove, Ashley Couch, Hanneke van Deursen, Chuck Driesler, David Fano, Silva Francisco, Jesse Ganes, Joe Geis, Andrew Heumann, Haruka Horiuchi, Olivia Howard, Florian Idenburg, Nicholas Kaspareck, Phil Kirschner, Adam Koogler, Jesse LeCavalier, Robert Linkens, Lauren Lynch, Jonathan Massey, Kyle May, Patrick McAndrews, Mick McConnell, Miguel McKelvey, Andrew McQuiston, Eric Moed, Paul Mosley, Corinne Murray, Federico Negro, Adam Neumann, Fed Novikov, Anibal Piccioni, Jacob Reidel, Lakshmi Rengarajan, Pierce Reynoldson, me, Michel Rojkind, Viral Shah, Sam Sidersky, Barbara Stallone, LeeAnn Suen, George Valdes, Devin Vermeulen, Robert Walker, Charles Weak, Dillon Wilson, Ebbie Wisecarver, and Phillip Youakim

I joined @Toroosian for episode 150 (!) of his Machine Minds Show to discuss the challenges of designing the interaction layer for a robot that could eventually do anything, and the challenges of keeping Rick from interrupting your podcast recording. Episode page: samsonrose.com/podcasts/desi… YouTube: youtu.be/q7m9tVxmgmU?si=KK95… Thanks to @ShakirWorks (@HelloRobo_co) for introducing me to Greg

$200M Series A, just... wow. Ryan Luke Johns has always been one of the most singular minds in construction robotics, and if you've seen my talks you know that earthmoving is one of the few task categories where on-site robotics makes sense for construction (when construction is industrialized, it becomes primarily offsite factory work and logistics). Congratulations to Ryan and the team at @GravisRobotics ! More info here: inc.com/georgia-fearn/exclus…

It's hard to overstate how revolutionary Claude Design is, not just as a design tool delivering unprecedented time-to-value to the designer, but as an omen of how all software interfaces will be irrevocably transformed

"Fully autonomous capture. Ground robotics and autonomous flight both get more investment." @procoretech to acquire @DroneDeploy after Drone deploy shrewdly acquired StructionSite and Rocos to build out their ground robotics product. Exciting news in the world of construction robotics, automated reality capture, and digital twins.

There was a small cadre of robotics designers at @figma Config 2026 but I suspect this group will grow. Software interfaces are the most obvious and tangible method of humans interacting with autonomous systems. The big idea from Config is that the design/build boundary is dissolving into a single canvas, with the agent(s) absorbing the work that can be standardized. There is no longer any meaningful distinction between an author persona and an end-user persona, and any application becomes capable of doing anything. This will be helpful with the application-scaling problem that accompanies the rise of general purpose robotics.

Designing the product that can do anything @figma Config 2026 youtube.com/watch?v=72m2hpP-…

cant believe I have to wait until 2026 for AGI

Predictably Spot-On Industrial Inspection | Boston Dynamics youtu.be/VVpgsd9Jsw0 Introducing Orbit fleet management and the Orbit site map

Stepping Up | Reinforcement Learning with Spot | Boston Dynamics youtu.be/Kf9WDqYKYQQ