How AI Learned to See Plants Better Than Humans

By Rinat Landman, Data Science at Agwa

Story from inside Agwa and the technology that understands every plant at every moment

When we began developing Agwa’s growing system, we quickly discovered a simple truth. In commercial agriculture there is one challenge no human can truly solve alone. Thousands of plants grow at their own pace, change every day and demand constant attention. As agronomists, we can carefully inspect dozens of plants, but scaling that level of care to hundreds or thousands across multiple locations is simply impossible. This realization became the foundation for Agwa’s work in artificial intelligence and it turned AI into an essential partner rather than only a useful tool.

Eyes That Never Get Tired

At Agwa, our system uses computer vision to continuously monitor plants from the moment they sprout until they are harvested. Small cameras capture images throughout the day and every Agwa device contains multiple plant spots. Each spot builds a complete visual history that shows the entire timeline of growth. But the system does far more than simply look. It understands, and it can

 • Recognize growth stages

 • Identify plant varieties

 • Detect early health changes

 • Uncover hidden patterns across multiple plants

An agronomist might notice a single wilting leaf. Agwa’s AI sees why that leaf is wilting and whether other plants began showing the same pattern earlier. Alongside vision, every Agwa device monitors its internal environment through real time sensor data measuring

 • EC

 • pH

 • Water temperature

 • Nutrient flow

These inputs help us detect issues long before they appear in images, such as nutrient imbalance, unstable irrigation or early water quality problems.

From Image to Insight

Every image inside Agwa’s platform moves through a multistage analysis pipeline.

First, the system identifies where all plants are, which spots are empty and what requires analysis. Then several specialized AI models run in parallel, each focusing on a specific task.

 • Plant growth stage

 • Plant identification

 • Health condition assessment

 • Nutrient deficiency detection

 • Wilting and stress analysis

 • Pest and disease symptom detection

Each model generates predictions with confidence scores. High confidence leads to automatic actions. Lower confidence triggers human review. Everything is designed to deliver accuracy at scale.

When Something Changes, We React

Agwa’s architecture is event driven. When the AI detects a growth stage change,

a developing issue or a plant that is ready for harvest, it generates an event that triggers the correct workflow.

 • Notifications to growers

 • Opening agronomy tickets

 • Database updates

 • Analytics and trend updates

The system also creates timestamped agronomy snapshots that record the full state of all plants at specific moments. These snapshots serve as the memory of the operation and help us track progress, compare decisions and validate performance.

Sensor anomalies such as a drop in EC, a rising pH trend or irregular irrigation cycles create device health events. These events may adjust operating policies automatically or escalate issues to agronomists before plant health is affected.

Decision Making at Scale

Agwa’s AI does more than detect problems. It makes decisions. It combines analysis outputs with historical data and generates

 • Recommended interventions

 • Severity levels

 • Prioritization

 • Predicted outcomes

When something requires attention, the system opens a structured ticket describing exactly what happened, how serious it is and which plants are affected. There is no need for humans to search for issues. They receive clear and actionable information.

Environmental sensor data is also part of the decision process. For example, a plant showing mild stress alongside a sudden increase in EC will generate a different recommendation than the same plant under stable conditions.

A Real Partnership Between Humans and AI

Even though Agwa’s platform operates autonomously, it works closely with human agronomists. Experts review decisions, resolve rare cases and provide corrections that help the AI learn. Each correction becomes new training data that improves accuracy.

The platform also tracks agreement rates between humans and AI, showing where the system performs well and where improvement is needed. Over time the AI becomes more confident, more precise and more aligned with expert expectations.

Built for Real World Agriculture

This is not a laboratory experiment. Agwa’s platform is a production grade system operating today in real growing environments around the world. It processes thousands of images and makes large volumes of decisions every single day.

The architecture is cloud based and fully scalable. It supports operations that would require many human agronomists to manage manually. The more data the system sees, the smarter it becomes. Hundreds of millions of sensor readings including EC, pH, water temperature and mechanical signals continuously enrich its understanding of every device and ensure reliability at scale.

Why It Matters

In modern agriculture, speed and scale determine success.

AI can

 • Detect issues earlier

 • Respond faster

 • Maintain consistent quality

 • Operate at massive scale

 • Learn continuously from every season

Humans excel at judgment, context and problem solving. AI excels at attention, consistency and volume. Together they create a system that makes growing operations more efficient, more predictable and more resilient. What once required constant manual effort can now be managed through intelligent automation supported by expert oversight. As Agwa’s system continues learning every season becomes more optimized than the one before.