Building From the Field Up.

The labor bottleneck is a system—not just an hourly wage.

What the public evidence shows

Seasonal labor is not a marginal input to German agriculture. Destatis counted 242,800 seasonal workers in 2023—28% of the agricultural workforce. In specialty crops, that labor is concentrated into short production and harvest windows.

A Bavarian LfL planning model makes an important customer distinction visible. Its wholesale field-strawberry scenario requires about 1,448 labor hours per hectare, including 1,125 for harvesting and 250 for manual market preparation. The comparable self-pick scenario requires about 385 hours. These are planning values, not survey averages, but they show that the same crop can have fundamentally different automation economics depending on how it is sold.

In a 2026 survey of 204 asparagus and strawberry businesses, the VSSE reported that many seasonal workers now stay for only four to eight weeks after reaching their income target. Labor can account for up to 60% of strawberry production costs. The constraint is therefore not only the hourly wage. It can also include recruitment, accommodation, onboarding, experience and the number of productive workers available during a narrow peak.

Mechanization does not automatically remove that system. A Bavarian horticulture research project concluded that hoeing robots are not yet generally economical and still have to be combined with manual hoeing. Automation often moves the remaining labor to quality decisions, preparation and finishing work.

How it changes the product

These sources do not show that every farm has the same labor shortage. Our fieldwork has also found farms with stable teams that return every year.

The useful target is therefore not simply a “labor-intensive crop.” It is a paid peak workflow that is hard to staff, concentrated in time, compatible with a bounded operating environment and not already removed by self-service or standard mechanization.

GroundMind should measure peak workers required, time to productivity for a new worker, saleable output per paid hour and human finishing minutes—not robot speed alone.

This does not make autonomous harvesting the first product. It narrows the next comparison: wholesale versus self-pick, field versus protected cultivation, new versus experienced workers, and picking versus preparation, sorting and crop care.

Paid peak workflow → bounded automation → fewer peak workers → verified saleable output

What comes next

We will test these distinctions in five to eight targeted grower interviews and on-site workflow observations. A workflow becomes a product candidate only when several independent farms can quantify the same bottleneck and at least one is willing to provide a timed observation, operating geometry or concrete trial discussion.

Grow strawberries, raspberries or another specialty crop in Germany? If you can show us one peak-season workflow, we would like to compare the public model with farm reality.

Public sources

• Destatis — 28% of Germany’s agricultural workforce were seasonal workers in 2023
• Bayerische LfL — wholesale strawberry planning model
• Bayerische LfL — self-pick strawberry planning model
• VSSE — Saison-Umfrage 2026, n=204
• LWG Bayern — economics and remaining manual work in horticultural hoeing robots

Even a one-centimetre pass depends on plant-by-plant judgment.

What we heard

A strawberry grower identified a recurring operation worth automating: removing weeds while loosening roughly the top centimetre of soil. The work remains manual because the right action varies from plant to plant, depending on the age of the strawberry stand, weed pressure and soil conditions.

The grower was open in principle to testing an early prototype on a small area, but immediately asked who would bear the cost of damaged plants or lost yield.

How it changes the product

The first workflow cannot be defined as “weeding” alone. It needs an explicit operating envelope covering crop stage, weed pressure, soil conditions, safe clearance and action depth. The system should act on routine, high-confidence cases and pause, skip or escalate whenever crop risk is uncertain. Crop damage is not only a technical metric; it is an economic and deployment constraint.

Plant context → risk assessment → shallow action → verification → skip or escalate → delayed outcome

What comes next

Observe and measure one complete manual pass before testing on live plants: area covered, task frequency, paid hours, current tools, soil variation and the existing baseline for plant damage. Any later trial should begin on a bounded test area with a comparison strip, predefined stopping rules and agreed risk allocation.

Harvesting creates the labor peak—but any first robot must prove crop-safe precision.

What we heard

Harvesting drives the labor peak: more than 40 pickers are needed in peak season, compared with around ten workers at other times. The grower would most like to automate strawberry and raspberry picking, but delicate fruit and uneven ripeness make crop damage and maturity judgment hard requirements.

Other recurring tasks may offer narrower entry points. Strawberry runners are removed by hand, while planting through plastic mulch requires every seedling to be placed precisely into its opening. The grower was open to testing a small prototype if it works reliably without damaging the plants.

How it changes the product

Keep weed treatment as the bounded first workflow. Use runner removal and precision planting as comparison cases for capabilities that should later transfer to more delicate work: perception, crop-safe action and verification.

Weed treatment → crop-safe perceive–act–verify loop → crop care → manipulation and harvesting

What comes next

Quantify runner removal and planting as future workflow candidates, while measuring crop damage, human intervention and labor saved in the weed-treatment MVP.

Farmers may buy completed field work before they buy a robot.

What we heard

A fellow early-stage agricultural robotics founder reported that conversations with vegetable growers changed when the offer moved away from selling machinery. Farmers were more receptive to paying for completed weeding work—similar to contracted labor—than to owning and maintaining a robot.

The same conversations suggested that robotic weeding is more credible as a repeated, early-season operation than as a rescue service for fields where weeds are already out of control. Direct-seeded vegetables may provide a clearer first test than crops where plastic mulch, canopy and close plant spacing increase the risk of crop damage.

This peer-reported learning from the United States closely aligns with signals from our customer discovery in Germany: the value lies in verified field work, not robot ownership alone.

How it changes the product

The robot should be treated as one component of a field operation. The offer must define the work, its boundaries and how completion is verified. Early trials should measure completed area, operator intervention, crop damage and net labor saved—not robot speed alone.

The service model should be tested through per-pass, task-package and seasonal-package offers before assuming that growers want to purchase equipment.

Defined operation → repeated passes → verified outcome → labor saved → seasonal service

What comes next

Test whether growers will move from general interest to a specific field trial with an agreed crop, operating window, success metric and price structure.

Automate the harvesting workflow, not necessarily the picking action.

What we observed

Berries require the most harvesting time because each fruit is small and demands repeated, careful handling. Labor shortage was not universal at this farm: a stable seasonal team returns every year, and transport is already partly mechanized through a simple remote-controlled platform. The harder problem is preserving saleable quality—avoiding bruising, defects and shorter shelf life.

How it changes the product

People may remain responsible for delicate picking and final quality judgment. A robot can reduce the work around that decision by following workers, moving crates, identifying ready areas, recording progress and escalating uncertain cases.

Observe crop → coordinate work and crates → record progress → flag exceptions → improve the next pass

What comes next

Measure saleable fruit per labor hour, walking and waiting—not fruit picked per hour alone.

Make narrow automation economical for small farms.

What we observed

An outdoor team of roughly six to eight people moves between harvesting, weeding and crop care. Vegetable fields should ideally be weeded every two to three weeks, adding up to several weeks of work over the year. Machines exist; the unresolved question is whether their cost and utilization make sense on small plots. In the greenhouse, tasks such as leaf removal are among the hardest for new workers to learn.

How it changes the product

A useful solution must move easily between plots and crops and earn its keep despite limited annual utilization. Success is measured through total cost, setup time, crop safety, reliability and labor actually released.

Small plot → bounded task → verified labor saved → reusable capability → more complex crop care

What comes next

Measure the real frequency, labor and setup burden of one bounded workflow before assuming a machine is economical.

Useful partial autonomy must produce measurable labor savings.

What we observed

Picking is highly time-consuming, and rising labor costs are putting pressure on strawberry economics. The grower named both weeding and picking as work worth automating, while stressing that weather, mud, irregular ground and crop variation make dependable outdoor operation difficult.

Automation on this farm is already collaborative: guidance and movement are mechanized while people still handle the plants. After years of practical experience with a field robot, the question is whether one machine can deliver enough reliable work to justify its cost across a short season.

The grower was open to a small future field test, provided the system can work reliably, protect the crop and prove useful labor savings.

How it changes the product

Start with a bounded workflow that is easier to verify than harvesting, and compare it with the farm’s existing machinery rather than with manual work alone. Let the robot perform repeatable actions; when plants are hidden, delicate or uncertain, it should pause, skip or hand the case to a person. Mobility, perception and verification should remain reusable across weeding, scouting, crop care and harvest assistance.

Rugged platform → bounded task → human escalation → verified labor saved → next workflow

What comes next

Observe a complete workflow, then define a small bounded test with agreed safety limits. Measure outdoor reliability, human intervention, crop damage, annual utilization and net labor saved—without treating conditional openness as a committed pilot.

Close one operation end to end before adding sensing complexity.

What we heard

Agricultural robots become useful when they fit the farm’s existing workflow. Technical performance alone is not enough: the product must close the full cycle—moving, finding, acting and verifying—at a speed that keeps pace with living, changing crops.

How it changes the roadmap

Start with one repeatable operation in a constrained environment. Use RGB-D for perception, localization and verification; add multispectral or local sensing only when it changes a farm decision and customers value the result.

Future field operations are designed to link what the system observed, where and when it operated, what action it performed, when a person intervened and what happened afterwards.

One operation → verified outcome → training example → better capability → next task

What comes next

Demonstrate one complete perceive–act–verify loop before expanding sensors, tasks or platform claims.

Field observations should become the next week’s work plan.

What we observed

Crop monitoring is part of daily work rather than a separate activity. Plant health is observed while people harvest, prune and move through the field. A more systematic inspection takes place on Mondays to plan the week, while labor demand peaks when spring and summer crops overlap.

How it changes the product

Automation should absorb extra workload during seasonal peaks and take on work that becomes repetitive or physically demanding over time. While executing a task, a robot should also record plant health, ripeness, anomalies and task completion without asking workers to enter the same information again.

Weekly plan → robot execution → observations during work → anomalies → task updates → next plan

What comes next

Test whether observations gathered during real work can reduce planning and documentation effort rather than create a second data-entry system.

Build for the point where existing tools return work to people.

What we observed

Weeding already combines mechanical tools and flame weeding. The interviewee estimated that roughly 30% remains manual; the workforce includes family, permanent and seasonal workers.

How it changes the product

The opportunity is not “automation where nothing exists.” It is the handoff point where existing tools stop working and return labor to people. Field work and its evidence trail—seeds, fertilizer, parcels and contractors—should be treated as one connected system.

What comes next

Measure the manual remainder by task, crop stage and plot structure, then identify whether the gap is frequent and valuable enough to support a service.