The robot moves down a row of infant carrot greens, its underbelly a continuous array of 232 tiny nozzles. It does not see the field as a farmer does, a blanket of green, but as a high-resolution map of thermal opportunity. In a fraction of a second, its vision system identifies a weed seedling, a speck of green peril nestled a quarter-inch from a crop plant. A micro-jet of superheated vegetable oil, precisely 232 degrees Celsius, strikes the target. The weed’s cellular structure cooks instantly. The carrot, untouched, grows on [Tensorfield.ag, May 2024]. This is the core interaction of Tensorfield Agriculture’s Jetty robot, a moment of extreme precision engineered to solve a problem of brute force: weeding.
The Wedge of Heated Oil
Tensorfield’s bet is not on a new class of herbicide or a stronger robot arm, but on a shift in the unit of agricultural intervention. Instead of spraying an entire bed or sending a crew with hoes, the company is selling millimeter-accurate thermal death. The technology, which the company calls thermal micro-jetting, uses computer vision to identify weeds and a proprietary manifold to deliver a pinpoint dose of heated, food-grade oil. The claims are specific: no soil disturbance, organic approval, and the ability to work in densely planted beds where mechanical tools cannot [Tensorfield.ag, retrieved 2025]. The initial wedge is the high-value, high-labor specialty row crop market, carrots, lettuce, spinach, where weed pressure is constant and the cost of manual weeding crews is a primary pain point. Tensorfield offers its system as a robotics-as-a-service, pitching a model of $50 per acre plus half a cent per weed killed, arguing one machine and operator can match the throughput of 40 people [AgFunderNews, ~2025].
A Team Built for the Hardware Grind
Founders, Backgrounds, and Bench
The founders assembling this system in the Bay Area are a classic deep-tech ensemble. The team is small and specialized, with over 40 combined years in robotics, computer vision, and mechatronics drawn from Stanford, Cambridge, and Imperial College [Emerging Ventures, 2026]. CEO Xiong Chang and CTO Cheehan Weereratne provide the AI and systems leadership, while co-founder Louise Thomas brings precision agriculture and mechatronics expertise [ZoomInfo, 2026]. Their backers signal a belief in hard tech development: SOSV’s HAX accelerator provides Shenzhen manufacturing connections, while other investors like Emerging Ventures and Broad Reach Ventures have placed early bets [SOSV, 2026]. This is not a software-only play; it is a bet that this group can navigate the long road from prototype in a field to reliable machines working commercial acres.
| Metric | Value |
|---|---|
| Jetty V3 Nozzle Array | 232 nozzles |
| Spray Precision | 0.25 inches |
| Target Service Fee | 50 USD/acre |
| Per-Weed Fee | 0.005 USD |
The Scaling Equation
Tensorfield is in the field trial phase, running commercial tests with a major West Coast produce grower [Tensorfield.ag, July 2025]. The ambition, however, stretches far beyond a single farm. In presentations, the company has outlined a goal of deploying five machines to reach an estimated $5 million in annual recurring revenue, with each unit having a potential ARR of $1.2 million [AgFunderNews, ~2025] [Embedded Vision Summit YouTube, May 2025]. The risks are not about the core technology’s ability to kill a weed, but about the operational and economic model holding up at scale.
- Deployment friction. Robotics-as-a-service in agriculture requires not just reliable machines, but a reliable service operation, transport, maintenance, and skilled operators in remote locations.
- Economic validation. The $50/acre + $0.005/weed model must prove cheaper than the alternative of manual labor or other robotic services across entire growing seasons.
- Competitive pace. While Tensorfield’s organic, no-till approach is a differentiator, well-funded rivals like Carbon Robotics are also moving down the cost curve and into new crops.
The company’s recent selection to pitch at FIRA USA 2025, a major ag robotics forum, suggests it is moving into a more visible phase of seeking partners and customers [Tensorfield.ag, Aug 2025]. The next twelve months will be about converting field trial data into signed service contracts, moving from technical validation to commercial proof.