Autonomous tractors are only the beginning. The next breakthrough in ag tech is a physical AI layer that can learn, adapt, and make farms dramatically more efficient without forcing every grower to become a robotics expert. Colin Hurd returns to the AgEmerge Podcast with a behind-the-scenes look at where ag autonomy really stands now, why retrofit solutions are only part of the story, and what has to change before self-driving machines become truly useful in the field.
From dusty harvest conditions that break today’s sensors to the opportunity for AI systems to preserve hard-earned farming knowledge, this conversation gets into the real bottlenecks – and the real upside – of the next five years.
Colin has devoted his career to leading innovative companies in the machinery industry. In 2016, Colin founded his second company, Smart Ag, which developed the first retrofit driverless system to automate tractors. After funding and growing the company, he led Smart Ag successfully through an exit to Raven Industries (NYSE:RAVN) in 2019. Raven was then acquired by CNH Industrial in 2021 (NYSE:CNHI). Colin stayed with Raven and managed business development until forming Mach in 2022.
Learn about Mach: https://mach.io/about-us/
Timestamps
07:08 – Why physical AI could transform the next 5 to 10 years on the farm
10:39 – Colin’s first startup and the soil compaction problem behind Track Till
12:06 – Why preventing compaction led him into autonomy
13:05 – Smart Ag, autonomous grain carts, and the Raven acquisition
14:34 – Building Mock by acquiring autonomy companies with complementary strengths
16:56 – Mach’s thesis: OEMs can source autonomy instead of building or buying it all
18:18 – Why off-road engineering talent is hard to recruit and even harder to localize
20:05 – Why lidar struggles in dusty ag environments
21:04 – Radar-based sensing as Mock’s bet for off-road autonomy
22:53 – Why OEM size creates different autonomy needs across large, mid-sized, and shortline brands
24:21 – Mach’s modular stack and why it stays OEM-only
26:20 – Why large enterprise users can help accelerate OEM go-to-market timing
27:29 – Retrofit autonomy and why Mach is avoiding that business model
29:14 – Why smaller, autonomy-first machine designs may be the real future
31:20 – Redesigning machines for autonomy instead of bolting it on
34:15 – One operator supervising eight machines in orchard autonomy
35:08 – Where humanoids fit into agriculture and what they would actually do
37:28 – From rigid path planning to non-deterministic physical AI
39:05 – Machines that understand objectives instead of only following pre-programmed routes
40:44 – The three core autonomy buckets: connectivity, situational awareness, and navigation
42:03 – Physical AI as a supervisory intelligence layer on top of the stack
43:03 – Agentic systems, self-repair, and the right-to-repair conversation
45:46 – Transferring farm knowledge, field history, and terrain awareness into machine behavior
47:16 – Using autonomy to support sustainable practices and reduce cost per acre
48:14 – Why cover crop adoption could accelerate if machines could learn from broader experience
50:37 – Physical AI as a way to reduce fear of failure in new practices
55:23 – Why recursive learning on-device is difficult, but data collection and fleet learning are realistic
58:32 – How farmers can connect with autonomy companies and stay current on emerging tech
63:07 – How Ag Startup Engine was built as an investor-managed fund with farmer input
65:39 – PlowVision and AI-guided cutting in meat processing
67:18 – Terra Blaster and real-time nutrient application for soil health
68:47 – Why founder feedback and customer validation matter more than hype
73:18 – The tradeoffs between angel investing and participating in a fund
74:13 – How farmers can get involved as accredited investors and pilot early autonomy tech
75:39 – Why Mach’s technology aims to make agriculture more efficient and more regenerative
