Jonathan Berte

Jonathan Berte's thinking is anchored in a single conviction that has held steady across nearly a decade: intelligence, once the exclusive product of human engineering, can now be manufactured. At SuperNova in 2018 he described the shift in almost theological terms, recalling that he used to believe "a system can never be smarter than its engineer" but that "this rule has been broken" ². For Berte this was not a rhetorical flourish but the founding insight behind Robovision — once you accept that "we no longer have to create intelligence our own, it's created for us... and the only thing we need to do is pay the electricity bill and invest in deep learning servers," the entire cost structure of building intelligent systems changes ². That reframing, from engineering intelligence to feeding and powering it, is the intellectual thread that runs through everything else he says.

From that premise Berte built an explicitly platform-first strategy, and he is candid about why. Robovision's early business model of custom image-processing projects failed to scale because it made the company itself, not the technology, the bottleneck: every new job meant assigning more smart people. The pivot to a self-service platform in 2014 was a direct response to that constraint, and Berte frames the ambition in the most maximalist terms available, stating flatly, "We want to become the SAP of AI" ². Tellingly, the platform's generality was validated empirically rather than assumed: a system built to teach machines to recognize plants transferred almost unchanged into solder-joint inspection in electronics manufacturing and into petabyte-scale medical imaging work at the NIH, which he cites as proof that the same underlying networks generalize across wildly different industries ².

Control and distribution are a recurring anxiety in his account of Robovision's growth. He describes a moment when customers began "trading USB sticks" of exported trained models, and he is blunt about what that meant: "we realized that we lost the control of the system, so we started to build our own AI store" ². He explicitly likens this to the way the Apple App Store consolidated the mobile disruption, treating distribution infrastructure as just as strategically important as the underlying models themselves ². This is a businessman's reading of a technical problem: the danger wasn't that customers were using the models, it was that Robovision no longer owned the channel through which value was captured.

Underneath the platform-building talk sits a more sweeping social ambition, one Berte states with startling directness. He connects the democratization of AI to the democratization of knowledge brought by the internet, arguing "if we can democratize expertise in the same way that we have democratized knowledge with the internet, we can build a better health care system for the world" ². He pushes this further into an almost political register, declaring "that is my true driver in life, is to kill the nation-state" ² — a statement that frames AI-driven democratization not just as a business opportunity but as a lever against centralized institutional gatekeeping, whether in healthcare, expertise, or governance itself.

By 2026, having moved from platform-builder to industry commentator with, in his words, "1,000+ machines deployed in the real world" , Berte's tone has shifted from evangelism to hard-nosed pragmatism, though the underlying economic instinct is consistent with his 2018 thinking about cost structures. Addressing the debate over general-purpose world models versus task-specific perception, he draws a clean line: "the romance is in general world models, but the economics are usually in task-specific perception" . He grounds this in operational reality rather than theory, noting that "compute isn't free, validation isn't optional," and illustrating the point with a deliberately unglamorous example: "a harvester in a greenhouse doesn't need to understand the universe — it needs to find the tomato, every time, at scale" .

What is notable is that Berte does not present this as a fixed ideological stance but as a moving target shaped by cost curves. He explicitly flags that "that line is moving," and identifies the real question as figuring out exactly "where world models start earning their compute cost" — a formulation that treats generality not as an end in itself but as something that has to justify its price. This is consistent with his 2018 framing of intelligence as something you pay for in electricity and server time rather than engineer by hand: the debate has simply moved from whether machines can be intelligent at all to how much intelligence a given task can economically afford.

Taken together, the throughline in Berte's public statements is a consistent refusal to romanticize AI even while being deeply invested in its expansion. He treats deep learning's arrival as a rupture ("this rule has been broken") ², treats distribution and platform control as inseparable from technical success ², treats democratized expertise as a route to reshaping institutions as large as healthcare systems and nation-states ², and treats the current wave of "Physical AI" hype around general world models with the same economic skepticism he once applied to bespoke engineering projects . The concrete takeaway for anyone building similar systems is his own stated discipline: match the model's generality to what the task and the compute budget can actually justify, because at scale — in his words, finding "the tomato, every time" — economics beats romance .