For decades, the crop input supply chain has run on a lag — growers, retailers and input companies reacting to a season already underway. Cropin CEO Krishna Kumar argues that’s starting to change. With Orbit AI, the agentic platform the company built on Google’s Gemini, Cropin is betting that AI can shift the input supply chain from reactive to predictive.
Cropin is working to turn insights that were previously invisible to agronomists into concrete, actionable guidance. Kumar says the technology can analyze field-level crop data and reduce blind spots in the input supply chain, give earlier visibility into harvest timing and yield, and flag quality-risks that affect sourcing and inventory planning
From Reactive to Predictive
Kumar points to input and seed companies as some of Orbit AI’s most mature use cases, citing work with BASF, Corteva, Bejo Seeds and Sakata Seeds across more than 30 countries. He says the AI tools can provide insights on planning to enter new geographic markets, disease pressures, climate effects, and more.
“For example, if you want to gain market share in a new region for a target seed, you’d want to know which pockets of that region are producing that crop, what past yield and disease pressure have looked like, so you can plan your market share,” he says. “They can decide whether to target 8% or 10% market share and how to grow year on year. And if they’re already in a region and want to map crop protection to a target crop, they want to understand which pockets have increasing or decreasing disease pressure by season, so they can plan inventory placement and have better conversations with growers.”
Cropin has also solved use cases for input players in seed production — matching supply and demand.
“Today the seed production industry does buffer planning of about 30% because anything can go wrong in a season and they don’t want to fall short. Can we reduce that by 10%, and what’s the impact on ROI?” he says. “Can I breed better seed based on the input I’m getting from the field? Shall I build a drought-resistant seed, based on the signals we’re getting — if there are certain diseases prevalent, or that region has less water.”
What This Looks Like for Agronomists
The same predictive shift plays out closer to the ground. In regions where one agronomist can be mapped to 40 or 50 growers, that’s more than any person can manage without help, Kumar says.
“One of the powerful impacts of this technology is in the hands of agronomists,” he says. “If you’re working with a CPG company with hundreds of agronomists partnering with growers to grow better and source better — those are the two things they want.”
He outlines the first thing Orbit AI does is highlight, based on risk — highest production and quality risk — how agronomists should approach their growers. The tool sorts which growers need attention due to water stress, poor crop health, and low nutrition uptake.
“They can just ask, ‘plan my day,’ which makes them very productive,” Kumar says.
From there, Kumar says, the platform surfaces two more issues that hit a food company’s bottom line directly: the size and quality of harvest.
“If they’re running short on the raw material, they check which farms in which part of the world are getting ready for harvest, and what next week’s harvest is going to look like — it runs a model called crop progression, which allows them to see these farms are ready to harvest,” he says. “They also get 60 days advance yield estimates, so they can plan their spot buying if they’re falling short, and plan their logistics, because — due to the climate — the crops are coming a week early, or getting delayed by 15 days because of a heat wave. So they have to plan ahead how to bridge that gap.”
Who’s Accountable When the AI Is Wrong
A predictive supply chain is only as trustworthy as the model behind it, and Kumar draws a clear line between the information provided by AI and what a human ultimately decides.
“For example, if we flag that nutrition uptake is lagging in certain patches, the agronomist goes and does the observation and suggests the dose — we don’t prescribe, we flag the issues,” he says. “So in the B2B, agronomy context, they decide the amount and dose. Our job is to say what yield you can expect — our accuracies range from 85% to 95%. We don’t over-claim on AI, because it can’t be 100% accurate.”
From Data Platform to Agentic AI
Cropin was founded a decade ago on the idea of bringing more data and intelligence to agriculture, starting with tools to help supply chains digitize grower networks and layer in agronomy, climate and disease-risk advisory. The company has since expanded to 103 countries, tracking 400 crops and 10,000 varieties, and says it has computed intelligence on a billion acres for its customers. It has raised roughly $47 million, most recently a Series D round that included a strategic investment from Google, and is headquartered in Bangalore with subsidiaries in the U.S. and Netherlands.
That data foundation is what let Cropin move quickly into generative and agentic AI, Kumar says. Between 2020 and 2023, the company compressed an open-source Mistral model into a lightweight “micro-LLM” capable of running on two-core GPUs, fine-tuned it on Global South agriculture data, and published it as open source. That groundwork led to Orbit AI, which runs on Google’s Gemini but is built on an MCP server architecture — meaning Cropin isn’t locked into one model and can deploy an open-source alternative inside a customer’s own environment if they don’t want their data leaving to train someone else’s model.
“The idea was to build something domain-specific for food and ag,” Kumar says.
