Happenings: AgTech Alchemy will be hosting an AgTech Alchemy: The Whole Hog event on July 22, 2025 around Tech Hub Live. We have 7-8 startups signed up and expect more than 50 people to show up.
Roombas and Rough Edges
In last week’s SFTW Plus edition, we looked at Amara’s law and how technology gets adopted. It is also important to compare any new technology to the existing best in class option available, whether it is a human or not, rather than expecting the new technology to be perfect. It is important to understand what is the benchmark to compare the new technology against.
This is especially true, when technology can be used as a substitute for some human capability like motor skills, or intelligence (with GenAI).
When the first Roombas came out, I was super excited that finally we would have a very clean house for little to no effort! We went out and bought one of the earlier versions of the Roomba.
We quickly realized for all the coolness of it bouncing around on the floor, there were many edge cases which it did not work really well with. For example, if you had a small object lying on the floor or had a room corner with some tight spaces, the Roomba would either get stuck or stop working because something got stuck in its cleaning mechanism.
So before starting the Roomba, one had to go around, adjust some furniture, close certain doors, and make sure there was nothing lying on the floor, which could disable the Roomba. Even before starting the Roomba, one had to make sure you had emptied out the trash holder on it from your previous run. The original Roomba’s very dumb. They would wildly bounce around like a drunk sailor. You hoped it covered 100% of your accessible floor area.
The newer Roombas (I use a brand called Shark) have solved some of the problems. Now you can let the Roomba map your house and you can instruct the Roomba to only clean the living room. You can schedule the Roomba to start at a certain time. The Roombas can now automatically go and empty the trash holder into another larger bin once it is filled. The battery life is better.
But you still have to prepare your room by picking up small objects and making sure the room is accessible which you want the Roomba to clean.
Any new technology has rough edges and we require humans to smooth out those rough edges to get the most out of the technology.
In fact, early on when the technology is still not mature and fully developed, you might actually end up spending more time and effort to make it work and give the technology developers an opportunity to improve it.
Rough edges with GenAI
One of the areas where GenAI seems to have the biggest impact is software engineer productivity. I have heard people say,
Do not study computer science, as your job will be eliminated. Everyone will be vibe-coding.
Many technology companies have reduced or stopped hiring new software engineers, and fresh computer science college degree holders have seen the largest negative impact on jobs in the last 2 years.
But a recent study (which could also be considered evidence of Amara’s Law), has presented information which debunks some of the claims of the impact of GenAI.
When developers are allowed to use AI tools, they take 19% longer to complete issues—a significant slowdown that goes against developer beliefs and expert forecasts. This gap between perception and reality is striking: developers expected AI to speed them up by 24%, and even after experiencing the slowdown, they still believed AI had sped them up by 20%.
Many startup technology founders have told me how they are able to iterate on their software quickly and efficiently.
Through my own experiments with vibe-coding, I can see the benefit of using vibe-coding to troubleshoot, get some basic stuff out quickly, and iterate. It does require you to have a good understanding of computer science fundamentals.
The idea that anyone can do vibe-coding is ridiculous. It has too many rough edges, which only a reasonable software developer can smooth out. The GenAI tools for coding become much more powerful in the hands of an experienced developer.
A monkey with a wrench can only wreak havoc, but an experienced worker can do amazing things with it. There are too many rough edges for a monkey to use a wrench effectively, whereas there are far fewer for a human.
Technology always has rough edges and we have to adjust our behavior to the rough edges of technology, to get the most out of a given technology.
When we say a young person is better at using some technology, some part of it is because they are better at smoothing out the rough edges of the technology. The genius of Apple and Steve Jobs with the iPhone was their ability to have as few rough edges as possible.
Currently, GenAI does have many rough edges.
It does not have access to unrecorded knowledge. It does not have the depth of understanding in a given field.
You have to ask a question in the “right way” to get the answer you want or expect. This has spawned the field of “prompt optimization”. Prompt optimization is nothing more than us smoothing out the rough edges of what a GenAI model can do.
There has never been a technology in the past, and if I can make a bold prediction, there will never be a technology in the future, which will not have rough edges.
Even amongst humans, we have different abilities to smooth out the rough edges of an existing technology. Due to this some people are masters at working with spreadsheets, whereas some can barely get two cells to add up with each other.
GenAI has definitely reduced the rough edges for many tasks as we can interact with the technology in natural human language, which all human beings have access to, though our skills with using natural language have a wide variance.
What does this all mean for AgTech?
GenAI is a powerful tool to help with many different workflows within AgTech. Do you have information and data which is unstructured, has high volume or scale, do the questions to be answered require creative thinking, personalization, and complex pattern recognition? These types of problems are very well suited for GenAI.
So some common use cases are answering customer service queries, helping a grower or farmer place a product effectively on their field (choose the right seed variety, fertilizer quantity etc.), help train your staff quickly and effectively, etc.
But even for all these use cases, and even for how amazing GenAI is, there are still a lot of rough edges which need to be smoothed out to actually get value from it.
For example, if you want to build a capability which helps your in-field team of agronomists and sales people to be able to respond to requests from producers and growers, you have to do many things and make sure they work together.
- Make sure your GenAI model has access to data needed to answer most of producer questions
- You have to do custom model fine-tuning
- Your data has to be in a format which GenAI can understand
- You have to make sure the answers are good enough
- You need to have deep domain knowledge of agronomy and the agriculture business
- Your team is on board to do business in a different way and so are your customers
- Your GenAI tools are integrated with existing workflows
- The experience has to meet user experience expectations
- You have to make sure you are following authentication and authorization, security, privacy, brand guidelines
- You have support from your management team and you are working on change management
- And any more.
As you go through this list, you realize your GenAI model is just one piece of the puzzle to deliver an experience and value proposition, which is actually useful to your users, customers, and stakeholders.
So there are still many rough edges to GenAI to use.
It still requires some old fashioned engineering work, it still requires a ton of domain knowledge, and most importantly it still requires a very deep understanding of your customer, including strong relationships and distribution.
You might say, where is the evidence?
OpenAI has quietly launched a services arm aimed at $ 10M+ enterprise clients, which includes forward deployed engineers, custom model fine-tuning, and multi-year contracts worth hundreds of millions.
OpenAI realizes that foundation models are important, but they are not the moat, when it comes to enterprise use cases.
The value derived from GenAI is very much dependent on how well you are able to smooth out the rough edges. We see this play out in almost every tool out there. Today almost 50% of workers say AI augments their work, but productivity lifts vary widely.
Smooth out the rough edges
The reason the productivity lifts vary widely is because of the difference in how different individuals and teams are able to smooth out the rough edges to get the most out of GenAI.
Every organization has its own rough edges when it comes to GenAI adoption (or the adoption of any new technology). Most of the organizational rough edges come from their data and technology infrastructure, the skill sets available within the organization and most importantly a hesitation to change.
Change management is the hardest part to help smooth out the rough edges of a powerful tool like GenAI and get the most value. As they say, learning is easy, but unlearning is hard. As human beings we fear uncertainty.
GenAI adopters are definitely pulling ahead, when they focus on clear use cases, have C-level sponsorship, assign and take ownership, and run fast feedback loops.
It is important to know that Big AI wins come from real use cases, real people, real domain knowledge, and strong change leadership.
Here is an easy check for your organization.
Does your CEO or your manager use GenAI in their day to day life?
If the answer is no, your GenAI project is most likely going to fail, as they have not displayed the basic curiosity for a business leader, and have not tried to work through the rough edges.
As an agribusiness or an AgTech startup, you have access to all the four ingredients needed to get the BigAI advantage over your competitors. You just need to figure out the rough edges and try to smooth them out.
What are you waiting for?
Do not forget to get my two white papers on GenAI adoption at the farm gate. The white papers provide a blueprint on how to get the most benefit out of GenAI capabilities with real case studies. They are available for free to all SFTW members.