The Bright Agricultural AI Future
The biggest clue as to how Generative AI use will develop for farmers is to look at what the rest of world is doing. Farmers have exactly the same opportunities to mould the new technology and so create the best form of agritech ever invented.
Non-generative AI precedes generative AI and study of leading non-generative applications delivers insights into the scale of the changes to come in farming.
Work on Waymo driverless taxis was started by Google in 2009 with the first fully driverless taxi service operating eleven years later in 2020. Waymo now operates in Los Angeles, San Francisco and Phoenix with other cities pending approval.
It goes without saying that driverless cars are at an extreme of the deep-trust-needed spectrum. Analysis of 3.8m miles of travel showed -76% accidents compared with human drivers (using the same roads and streets) and zero human injuries. Clearly, these data speak for themselves.
Business-critical, but with less of the life-and-death risk element has been Amazon’s expansion of AI robot use. From an acquisition in 2012, Amazon in 2025 was deploying 1m+ robots of eight kinds including one bipedal (walks like us) robot, all working alongside, at the moment, approximately the same number of human colleagues.
The Big AI Difference
The development of Generative AI had been quite quietly happening in parallel with non-generative AI but in 2017 there was a breakthrough discovery (ironically, made by a Google team) leading to the public release of ChatGPT by OpenAI in 2022. Such is the sheer functionality of Generative AI it is clearly heading for saturation of usage within a few short years, almost certainly by 2030.
Most significantly for users, and unlike driverless taxis and warehouse robots, there is zero or very low capital expenditure required by the user. Equally significantly the rate of evolution of generative AI capabilities has been quite astonishing. OpenAI alone, excluding all their competitors' many alternative products, has released ten major upgrades since the first public ChatGPT. Sit tight...
The pace of model advances is breathless. A common occurrence now is for academic research into the application of AI to be published on work based on a version half-a-dozen or so advances behind the model current on the paper's publication date.
Capability Overhang
A Capability Overhang may sound like a landscape feature to be found in craggier parts of some of the deer parks surrounding the great houses of Britain. It actually refers to the gap between all that Generative AI is capable of and what has been uncovered so far.
Another way of scoping the difference between discovered and un-discovered capabilities rests in the putative hypothesis that ‘if all AI development was put on pause now, it would take five years for all its possible uses to be uncovered’.
But of course development is racing ahead and thankfully, and counter-intuitively, the more powerful Generative AI becomes the easier it is to use: the greater power further helping to democratise knowledge for all, including inexpert users.
A particular example was the need for AI prompting skills: users once needed these to get the most out of their AI . . . that is until the release of the o1 model in September 2024, an event that shifted Generative AI on its axis.
The only initial ‘prompting’ required with the o1 kind of AI, known as the reasoning models, is to tell the AI what you want to do and maybe why you want to do it.
For farmers the benefits of using Generative AI continue to grow apace with every system upgrade: greater capabilities combined with simpler and easier useability.
The Really Difficult Bit
The hardest thing to know about Generative is, given the massive leaps in performance over the first three years coupled with the complete lack of any signs of the rate of increasing performance slowing down, what it will be capable of even this time next year - ignoring where it will be in three, five and ten years.
Our AI future is incandescently bright.