Robert Thelen has a blunt thing to say about how most companies are currently building with AI, and he's been saying it in rooms where people don't always want to hear it.
If your product runs on a frontier model, someone else owns your competitive advantage. Full stop.
"There's nothing proprietary about it," he told me when we talked recently. "Your code can be easily copied, your product can be easily copied, and you're using a model that everyone else has access to." The frontier model providers, he argues, are watching which API use cases get traction. And then they build that feature themselves. It's not speculation. It's a business model.
Robert is CEO and co-founder of Rownd, a company that is moving AI from the cloud to the front lines. Rownd provides the tools, apps, and frameworks to fine-tune specialized models, train vision systems, and run anomaly detection directly on the edge — meaning on your own hardware, disconnected from the cloud if needed. By fusing real-time sensor data, vision, and local knowledge with agentic frameworks, they've moved well beyond chatbots to build autonomous agents that control hardware and move things in the physical world.
To make sure that the future stays accessible, Rownd actively maintains LlamaFarm, an open source project that gives anyone the foundation to build smarter-than-the-cloud systems that collaborate across local compute. The project recently crossed 824 stars on GitHub and is free to use. If you want to dig in, the repo is at github.com/llama-farm/llamafarm, and the community lives on Reddit.
The use cases Robert is focused on are the ones where connectivity can't be assumed, and failure isn't an option. Defense. Healthcare. Edge hardware. Drones. If a hospital's cloud provider goes down for an hour, patients still need AI-assisted coverage. If a drone is operating in an environment where satellite connection is being actively jammed, it still needs to function. These aren't hypothetical edge cases. They're the reason Robert left a comfortable path at IBM and went through Y Combinator to build this.
The origin story matters here. He spent eight years in the Air Force: aircraft maintenance, then a year in Afghanistan as a combat advisor training Afghan army forces, then acquisitions work. Not a traditional tech pipeline. But the through line is learning to adapt fast in environments that don't give you a manual. That skillset traveled. After his MBA at Dartmouth's Tuck School of Business, he interned at Google, landed at IBM as a Senior Product Manager, and eventually convinced his co-founding team — Matt, Rachel, and their first employee Bobby, all IBM colleagues — that there was something worth building outside the walls.
They got into Y Combinator in 2022. They were building an authentication product, doing well, had crossed $2M ARR. And then they watched their own category start collapsing in slow motion.
When Robert's team started using Claude, they realized it could replicate what their authentication product did, well enough, cheaply enough, that customers would eventually leave. Not right away. But the curve was clear. So before anyone asked them to pivot, they did it themselves. They killed the working product and moved into edge AI infrastructure before there was a crowd there.
Now there's a crowd coming. And Robert wants to make sure the people building in this space understand what it means to actually own what you're building.
The talk he's giving at All Things AI is a specific example of that philosophy in action: how to use open source tools to autonomously control drones. Not a concept demo. Real hardware, real models, running locally.
"The coolest thing about All Things AI," he said, "is that the speakers are people who build models, build frameworks, build agents. They're not just talking about ChatGPT."
That's the room he wants to be in. And honestly, it's the room worth being in.
Robert's founder journey is also featured in our Raleigh-Durham Startup Week spotlight series. RDSW runs April 20–24 across Raleigh and Durham