There is something pretty remarkable happening in AI development right now. Work that once required weeks of infrastructure setup, pipeline integration, and specialized engineering can increasingly be done in an afternoon.
The interesting part is what’s making that possible. Coding agents are getting much of the attention, but the real acceleration comes when those agents have access to a complete, functioning data and inference stack.
That’s the idea behind the VAST Builders Challenge, a series of hands-on, single-day events where developers build working video AI applications using Cursor, NVIDIA models, CoreWeave GPU infrastructure, and VAST AI OS. Rather than spending their time configuring storage, moving video datasets around, or wiring up inference pipelines, participants arrive with the heavy lifting already done. They can focus on building something useful.
In this episode of Shared Everything, I talk with Brian Verkley, Director of AI Data Platform at VAST Data, and Ram Bansal, Senior Developer Advocate at VAST Data, about what happens when you remove those infrastructure barriers and give developers access to tools and datasets they might never have been able to work with otherwise.
The technical stack is worth a closer look. VAST AI OS handles video storage, ingestion pipelines, contextualized data, and the application services, while NVIDIA’s Cosmos reasoning model runs on CoreWeave GPUs. Cursor provides the coding agent interface, with skills that let developers interact with the underlying platform, search video, change contextualization prompts, and build new application functionality.
It’s a practical example of how the developer’s job is shifting toward defining what an application should do rather than manually implementing every component.
And the results have been surprisingly varied. Participants have built applications that identify gaps in robotics training datasets, analyze cycling footage for road hazards, automate infrastructure maintenance reporting, and manipulate video through natural-language instructions. Some of these projects would previously have required substantial development time. Here, they were functional within hours.
One of the more interesting points Brian and Ram raise is what happens after that first successful prototype. Moving toward production means expanding datasets, testing more scenarios, evaluating nondeterministic outputs, and ensuring that applications behave reliably beyond the handful of examples used to build them. Faster development doesn’t eliminate those engineering challenges, but it does change how quickly teams can get to them.
The bigger takeaway is that AI development is moving toward a new abstraction layer, where coding agents, models, and data infrastructure work together as a builder stack.
The Builders Challenge offers a useful glimpse of what that looks like in practice, and why the next wave of AI applications may come from a much broader group of builders than we’ve traditionally expected.
If you missed this round, keep an eye on exciting developer events happening at VAST FWD 2027.
