The Farm as a Stack

Every generation of production systems eventually demands a new generation of infrastructure.

The earliest farms were infrastructure-light: land, labor, seed, rain. Over time, we added irrigation, mechanization, synthetic inputs, GPS guidance. Each addition created new dependencies — and new leverage.

Now something different is happening. The infrastructure itself is beginning to act.

What "agentic" means

Agents are not software that waits to be asked. They are software that watches and responds.

A traditional dashboard shows you soil moisture. An agent notices soil moisture falling faster than expected, checks the forecast, sees no rain for a week, and sends you a message: "Block 12 will reach stress threshold by Thursday if not irrigated."

The difference is initiative. The dashboard presents. The agent anticipates.

This matters because attention is finite. A grower managing thousands of hectares cannot monitor everything. The value of an agent is not that it knows more — it's that it watches while you work on something else.

The farm is becoming a stack

In software, a "stack" refers to the layers of technology that work together to deliver a product. There's infrastructure at the bottom, platforms in the middle, applications on top.

Agriculture is developing its own stack. And like any stack, it has layers that depend on each other.

Layer 1: Infrastructure for agents to act on.
Sensors, controllers, connectivity, machinery APIs. The physical substrate that makes digital action possible.

Layer 2: Infrastructure for building and running agents.
Data pipelines, model hosting, orchestration frameworks, integration layers. The platform that makes agents possible.

Layer 3: Infrastructure that is itself agentic.
Systems that watch, decide, and act — with appropriate human oversight. The intelligence layer.

Most farms today have fragments of Layer 1. Very few have coherent Layer 2 infrastructure. Almost none have Layer 3.

This is the opportunity.

The connectivity problem

Before agents can act, they need access. Access to sensors. Access to weather data. Access to imagery. Access to machinery controllers. Access to historical records.

Today, this data exists in silos. The weather station talks to its own cloud. The irrigation controller has its own app. The tractor logs to a manufacturer portal. The agronomist keeps notes in a spreadsheet.

An agent cannot reason across systems it cannot see. The first infrastructure challenge is simply: connection. Bringing the relevant data streams into a unified space where reasoning can occur.

This is less glamorous than AI. It is also more important. A brilliant model with no data access is useless. A simple model with complete data access can be transformative.

Who owns the recommendation?

When an agent suggests an action — irrigate, spray, harvest — that recommendation carries weight. It influences decisions. It shapes outcomes. It affects profit.

Who is accountable when the recommendation is wrong?

This is not a philosophical question. It is a design constraint. Agentic systems that operate on farms need clear boundaries around autonomy. They need explicit guardrails. They need audit trails.

Our view: the agent proposes, the grower decides. Autonomy should be earned incrementally, through demonstrated reliability, with human oversight always available.

The goal is not to remove human judgment. It is to make human judgment more powerful by providing better information, faster.

Building the agentic layer

At frukht, we build custom AI agents for agricultural operations. Each agent is tailored to a specific context: crop type, geography, infrastructure, management style.

The work begins with connection — integrating the data sources that matter for that operation. Then comes modeling — building the reasoning layer that translates data into insight. Finally, action — creating the interfaces through which recommendations flow.

The result is a system that watches, learns, and helps. A system that gets smarter with each season. A system that remembers what worked and what didn't.

The stack is being built. The question is who will build it, and for whom.

The agent proposes. The grower decides. That is the design.