Exploring AI with Bushel

On the engineering front, it’s pretty important to us. We’re using Copilot, which is by Microsoft. It helps engineers write better software and more efficiently. Our general thesis is that it makes an engineer 20- 30% more efficient. In some cases, it’s as capable as an intern or a junior developer sitting alongside you.

We started using Copilot in the middle of the year last year via a test case and that went well enough that we included it in our budget this year to have that tool for most of our team. It has definitely been worth the time and money.

Q: Were you hesitant at all about adopting these tools?

A: Personally, I am biased towards using these tools. Although you don’t necessarily want to be putting proprietary information into ChatGPT. Companies like ourselves, which have IP and private information, need to make sure they’re being smart about where they’re putting their information.

When we did the test case last year with Copilot, we read through and worked on what we could and couldn’t use Copilot for. We worked on figuring out how they were using our information and at the end of it, we felt comfortable providing it to our team.

Did You Know? Bushel completed a $47 million Series C investment fund in 2021.

Q: How much training did you have to do around that in making sure that your employees weren’t putting sensitive information into the AI systems?

A: We just made sure were clear about our expectations and what we were and weren’t comfortable using it for. It really hasn’t taken up a lot of our time.

Q. Are there any other tools you are using?

A: We’re currently developing some Custom GPT models for our team and our customers using our data. The idea is to create what we’re calling a Bushel Buddy Bot. We taught it how to read our documentation and how to understand any questions our own team may have with our internal documentation at Bushel.

For our customers, we’re helping answer questions like if a farmer wanted to know how much corn they delivered last year versus the year before that. This would also work on the other side for retail customers that we have as well.

We’re doing this using mostly Microsoft tools and partially through OpenAI’s private business tools as well—we feel more comfortable doing proprietary work with this.

Q: It should like you haven’t done anything yet with that unique data set that you mentioned earlier, is that correct?

A: We certainly haven’t built our own models yet using heavy AI tools. We do think we have a pretty unique data set though. We’re still trying to figure out the use cases for our customers. We have to focus on the right ways we can help the customer and we also have to focus on what is a good use of their information and what is not. There are things that are on the balance of being inappropriate or things that we can’t do given our agreements. So, we have to be smart about how we think about solving any particular challenge for the customer.

Q: Can you tell me more about those challenges?

A: Well, in any kind of business case where you have customer data, it’s proprietary to the customer, and, generally, you’ll have agreements that will dictate what you can and can’t do with the information. So firstly, you need to follow those agreements really well. The second part is even if you have the right to do certain things, you have to decide if it’s beneficial and who it is beneficial for. You could imagine having information and tools available to somebody who is interested in it. But agriculture is a twosided market—there’s always somebody buying and always somebody selling. And if you’re only providing value to one side of that equation, then you’re creating a negative effect on the other side, which could be our customer. So, we have to be very smart about which capabilities we’re pursuing that would actually help our customer because we have no interest in hurting our customer. So, that’s a trick to balance.

Q: What are some improvements you would like to see with AI going forward?

A: ChatGPT’s models aren’t necessarily good at math or understanding metrics without context. For companies like us, being able to understand numbers and do calculations with numbers is more important than the actual language itself.

I think the day when it’s solving real problems is when we no longer have to build business dashboards or reports. These tools should be able to do that at some point down the line— that will be a good leap forward. I think we’re pretty close.

Q: How does Bushel stay up to date on recent technological advancements and know when to seriously look into something?

A: That’s a challenge. You can’t just sit back and hope it comes to you. You have to be very proactive. We have a few people who are tasked with owning this as part of their job for Bushel. They make sure we’re looking at these kinds of tools and capabilities and trying to find ones that could disrupt or make our business better.

You have to be interested and willing to self-educate. We have a lot of people here at Bushel who are curious and using various tools themselves. That helps with a lot of ideas being brought to the table.

Q: Is there anything else you want to say to our readers?

A: If you remember a few years ago, blockchain was the coolest thing since sliced bread—it was going to change every single thing in our lives and solve all of the world’s problems. If you look today, there hasn’t been a lot of real business impact to come from blockchain. It certainly hasn’t changed agriculture.

But, to think that AI is going to be the next fad would be a mistake. AI’s development and advancements are more akin to what happened with the internet. It’s going to fundamentally change how we do business, how we build software, and how we do our jobs. The changes over the next 5 to 10 years are going to be so incredible that we can’t comprehend them right now.