An Operating Model for AI-Native Tropical Agriculture
Where Does Philippine Coconut Farming Stand Today?
Philippine coconut-sector sources commonly report roughly 2.5 million coconut farmers and approximately 3.6 to 3.65 million hectares under coconut cultivation — a land base second only to Indonesia. As many as 25 million Filipinos are often described as directly or indirectly dependent on the sector. The economic centre of gravity in the global coconut trade has, for most of the post-war period, sat in this country.
Commonly cited estimates put smallholder yields somewhere between 0.5 and 1.0 tonnes of copra-equivalent per hectare per year. Best-managed plantations under continuous capital investment reach 2.5 to 3.5 tonnes; experimental sites with full agronomy support and modern hybrids touch 4.0 tonnes. The gap is not a few per cent but a multiple. The technical frontier is also moving — modern genetics, intercropping science and precision nutrition mean the achievable yield in 2030 will be higher again — but the smallholder average has barely moved in fifty years.
Behind the headline numbers is a harder truth that does not appear in most policy documents. Coconut palm ownership has fragmented across generations and is often held jointly within extended families. By itself that ownership stake rarely produces an income that can support any single household member full-time. The actual day-to-day work — climbing, harvesting, husking, hauling, drying — is done predominantly by tenant farmers and farm workers under the “kasama” sharecropping arrangement that remains widespread. The owners are typically not the farmers. The farmers are typically not the owners. Formal tenancy and informal share-work arrangements remain significant in many coconut-growing areas, although precise current shares vary by locality and source.
What this means in practice is that coconut is, in many parts of the Philippines, no longer a single meaningful occupation. Many rural households combine coconut income with wage labour, fishing, livestock, small trade and remittances from family members working overseas; in such households, coconut income is often only one component of a wider survival portfolio. The pattern is sector-observed and field-observed rather than precisely measured at the national level for coconut specifically. The big question that follows is the one we keep coming back to: where will the next generation of coconut farmers come from?
The reasons the system arrived here are well understood by anyone inside it. The Philippine coconut estate is dominated by senescent palms; the median farmed palm is fifty to seventy years old, past the productive plateau, on a downward yield curve. Replanting is economically untenable for smallholders because the new palm takes five to seven years to come into bearing. Inputs are under-deployed, pest and disease pressure has been chronic, intercropping is practised on perhaps a fifth of farmed area, and the post-harvest chain captures most of the value at a price set by the prevailing CNO benchmark in Rotterdam — with limited transparency and almost no premium for quality, sustainability or traceability.
The result is a structurally low-productivity, low-margin, low-resilience system that is also presented to the world — in marketing materials, development reports, supermarket shelves — as a sustainability story. The dissonance between the lived economic reality of the coconut farmer and the sustainability narrative around the product is the single largest moral and strategic problem in the sector. It is also the problem this paper sets out to solve.
Why Is the Productivity Gap the Poverty Gap?
The income arithmetic is worth doing explicitly, because the abstraction of “productivity” obscures a household-survival equation, and the multi-occupation reality changes how the arithmetic should be read.
A coconut smallholder operating two hectares at a yield of around 0.7 tonnes of copra-equivalent per hectare per year — a value within the commonly cited 0.5–1.0 tonne range — produces roughly 1.4 tonnes of copra annually. Sold at the prevailing farmgate price (over the past five years averaging between USD 550 and USD 750 per tonne), the gross income from coconut on those two hectares is in the order of USD 770 to USD 1,050 per year. Net of inputs, harvesting labour, transport and post-harvest losses, the household retains perhaps sixty per cent of that — a coconut income of roughly USD 1.30 to USD 1.70 per day for a household of five. In coconut-dominant regions of Mindanao, Bicol and Eastern Visayas, poverty levels are commonly reported above the national average.
Coconut-farming households are consistently identified in Philippine policy literature as among the poorest agricultural groups in the country. It also explains why the children of coconut farmers do not, on the whole, become coconut farmers. And it explains why, for the worker who climbs and harvests the palms, coconut is rarely the only occupation: two or three other jobs — wage labour on neighbouring farms, fishing, motorcycle-taxi work, small-scale trade — are typical. Each is a piece of the patchwork that lets a household survive month to month.
The strategic point is that the productivity question and the “single meaningful occupation” question are the same question. The minimum yield, margin and price stability at which coconut farming becomes a viable single occupation is the productivity threshold. Below that threshold, coconut is necessarily one job among several. Above it, it can support a household on its own and begin to fund the reinvestment the next generation requires. Closing the productivity gap is not merely income arithmetic; it is the difference between a sector that exists as a residual livelihood and a sector that exists as an industry. Lionheart’s working definition of success is the second of those.
A sustainability story alone will not fix this. The premium that organic, regenerative and Fair-Trade-aligned certifications attract — which Lionheart has captured — is real and meaningful, but the order-of-magnitude problem is not a fifteen per cent premium on a low-yield system. The order-of-magnitude problem is a low-yield system. At three tonnes per hectare instead of 0.7, the same premium would be transformative. At 0.7, no premium is transformative.
The conclusion is operationally clear: in coconut, sustainability is downstream of productivity, and productivity is what allows the work to become a single, dignified, full-time occupation again. A farmer who is not solvent cannot afford to be sustainable. A farmer who is not productive cannot be solvent. And a sector in which farming is one of three part-time jobs cannot retain the next generation.
How Does Productivity Become Sustainability?
The thesis can be stated in one line: productivity is sustainability, because productivity is the only variable in the sustainability chain that, when solved, makes every other link feasible.
The chain is short. Productivity drives gross margin. Margin drives equitable income. Income drives household reinvestment. Reinvestment drives continuity into the next generation — children who study, who go to university, who choose to come back to the farm or to the agri-sector because there is something there to come back to, including the chance to make coconut farming a single, dignified occupation rather than one part-time job among several. Each link presupposes the one before it. Cut any link and the system collapses into dependency.
The strong corollary, sometimes uncomfortable, is that sustainability commitments by themselves do not break the dependency cycle. Certifications, fair-trade-aligned premiums, regenerative-practice protocols, ESG reporting frameworks — all are necessary conditions, and Lionheart maintains them rigorously. None is a sufficient condition. The sufficient condition is a productivity uplift large enough to pay for everything else.
What changes the picture is not a new certification scheme or a new donor programme. What changes the picture is a structural increase — potentially a step-change in management density, with order-of-magnitude effect in selected workflows where the baseline is highly manual and fragmented — in the management density and operational granularity we can apply to a smallholder farm. That is what AI-native operating models, on top of a stack of complementary technologies, make possible for the first time. It is, in the most literal sense, what “productivity is sustainability” means in practice.
Why Can't Traditional Management Scale Smallholder Farming?
Smallholder tropical agriculture is a high-variation, low-margin, geographically distributed system. Each farmer is different. Each parcel is different. Each palm is different. Each season is different. The variability is biological, not industrial — it does not lend itself to the standardisation logic that built modern manufacturing. Yet for fifty years the dominant management response has been to standardise: one fertiliser protocol, one spacing recommendation, one harvesting cadence, applied across millions of hectares of structurally heterogeneous farms.
Standardisation is administratively convenient. It is also yield-destructive — and unfair to the worker. A standardised fertiliser regime under-supplies the high-demand farms and over-supplies the low-demand ones. A standardised harvesting cadence misses optimal nut maturity on most palms most of the time. A standardised work-norm penalises the worker assigned to the harder palms and rewards the worker assigned to the easier ones for reasons that have nothing to do with effort or skill. The result is a yield distribution dominated by its low tail, and a workforce that experiences the system as arbitrary.
The alternative — bespoke management per farmer, per palm, per soil pocket, per harvest condition — has been understood since the early colonial-era plantation studies. In practical terms, the factor-three claim is the difference between prevailing smallholder yields of roughly 0.5 to 1.0 tonnes copra-equivalent per hectare and well-managed systems capable of roughly 2.5 to 3.5 tonnes per hectare; the uplift should not be attributed to management alone but to the combined package of management density, genetics, nutrition, pest control, replanting and harvest discipline. Done well, the change also dignifies the work, because the worker’s effort and the work’s difficulty become legible.
The reason it has not been deployed at smallholder scale is not a failure of insight; it is a failure of arithmetic. To deliver bespoke management at the country’s coconut estate, the Philippine sector would need a professional agronomic and field-management capacity far larger than it currently has. The actual number deployed is a small fraction of what bespoke management would require, and the public extension service is structurally under-funded — one extension agent per two to five thousand farmers is typical.
Even on the largest, best-capitalised plantation companies in tropical agriculture, the management-density wall is real. Once a company exceeds a few thousand hectares, per-palm decision quality degrades, no matter how many agronomists it hires. The wall is not an industry failure; it is a mathematical one. Human management does not scale linearly with biological variation, and biological variation is what coconut is.
What Does AI Change?
Until now. What artificial intelligence — specifically, the new generation of agentic, multimodal, edge-deployable AI — combined with the broader technology stack of sensors, drones, mobile devices and increasingly capable field robotics, changes is the unit cost of management itself. A trained agronomic AI agent, deployed on a phone, advised by a small human team, can in principle hold the productivity-relevant state of every palm in a portfolio in working memory and act on it in near-real time. Management density that was financially impossible at any scale is now technically routine. That, and not anything else, is the point of inflection the sector has been waiting for.
It is worth being precise about what we mean by tech and AI as a management multiplier. We do not mean a chatbot bolted onto an existing process, or a forecasting tool consumed in dashboards by head office, or the automation of a single task. What we mean is a fleet of domain-trained AI agents, each with a defined mandate, running continuously inside an operating company on top of a stack of complementary technology — sensors in the field, drones for canopy and yield estimation, mobile devices in the hands of field operators, traceability-grade scales and ledgers in the post-harvest chain, and increasingly task-specific robotics. The agents observe, decide, act and report; the underlying tech stack supplies the eyes, the ears and increasingly the hands. The boundary between agent decisions and human decisions is explicit and is set by the company’s governance rules. High-stakes decisions stay human. Critically, the time saved by the agents is reinvested deliberately into more human contact — more handholding, more coaching, more farmer-facing presence — rather than spent on overhead reduction.
What this changes structurally is the relationship between management quality and headcount. Before AI agents, every additional unit of management quality required an additional unit of headcount; the productivity ceiling of a farm or a brand was bounded by how many agronomists or e-commerce managers a company could afford. With AI agents in the operating layer, a doubling of management quality may require no additional headcount, because the marginal agent costs almost nothing to run. Headcount is replaced, in part, by model quality, supervision quality and the granularity of the underlying tech stack. The implication for tropical agriculture is direct: the constraint on smallholder productivity has always been management density, and AI-native operating models can deliver a step-change in management density — potentially an order-of-magnitude improvement in selected workflows where the baseline is highly manual and fragmented.
What Does This Look Like in Rizal, Palawan?
The first of the three live demonstrations at the FEU keynote on 28 April 2026 was the Harvest Productivity Agent, a workforce-planning and performance-evaluation agent currently in early-trial deployment with our Track One sap operation in Palawan.
To understand what we have built so far, it helps to picture the work the agent serves. Before the sun fully rises on a typical morning at our Rizal farm, a group of workers we call tappers begin their day. Their job is to climb coconut palms — short ones, tall ones, hundreds of them across a workforce of more than a thousand active tappers — cut into the unopened flower bud, and collect the sap that drips out. Sap does not wait. There are specific windows in the day during which the bud has to be cut, the tap made, the collection vessel emptied, or the yield is lost for that palm on that day. The operational question is daily and unrelenting: who goes to which palm, in what sequence, in which window, given each tapper’s speed, skill, and the conditions of their assigned area.
Pre-agent, that question consumed a recurring four-person, four-hour daily commitment from the planning team. They reconstructed yesterday’s work, evaluated the count of palms tapped against the count expected, identified gaps, and built today’s assignments by hand. Even with that effort the result was approximate, because some tappers were faster, some more consistent, some newer to the work, and assigning them by hand introduced variation that had nothing to do with their performance. Underneath the planning sat an evaluation problem: how do you fairly grade a tapper who works a difficult cluster against one who works an easier one? Pre-agent, you could not — at least not with the consistency that fairness in operational work requires.
The agent we have deployed addresses planning and evaluation together. It looks at each tapper’s record — palms covered, consistency over time, output against a normalised standard for the conditions they were operating in — and grades each tapper on the same set of criteria, every day. The grading then drives the next day’s plan: who covers which palms, in which sequence, in which window. The agent surfaces three things to the supervisor on a single screen: the top performers, the patterns of underperformance broken down by tapper and cluster, and the distinction between effort that is genuinely lacking and conditions that are genuinely difficult. The supervisor’s job is then to coach.
The four people, four hours a day previously absorbed by planning and evaluation can now be reinvested. They are reinvested, deliberately, into more time with the tappers themselves — coaching, training, accompanying, problem-solving. This is the handholding principle of Chapter 9 in operational form: the agent handles the cognitive grind that produced no community value; the supervisors spend the recovered hours on the human touch points that produce all of the community value. The yield consequence — more palms tapped at the right windows, with sap collected before it is lost — is the second-order effect of getting the first-order operational discipline right.
Want to see the farm and community model behind this? Explore Lionheart's approach to regenerative farming and community.
Where Will the Next Generation of Coconut Farmers Come From?
If the architectural argument of this paper is correct, the most consequential implication is therefore this. The big question — where will the next generation of coconut farmers come from? — does not have an inevitable answer. It has the answer that we, as a country and as a sector, decide to build for. The next generation of Philippine farmers, owners, supervisors and operators can look nothing like the last. They can manage technology, not just trees. They can own equity in cooperatives, not just provide labour to them. They can build careers in coconut as a single, dignified, full-time occupation rather than as one part-time job among several. They can be better educated, better connected, and better paid. None of that happens automatically. It happens only if companies like Lionheart, the wider ecosystem of operators, investors, academics and government partners around them, and the next generation of students themselves succeed in building AI-native agri-operating models at meaningful scale, on the Our Farm to You principle that connects host communities directly to the consumer who pays for the work. The opportunity is real and it is tractable. The work is in front of us.
What Is the Work in Front of Us?
If the productivity gap in Philippine coconut is the poverty gap, the demographic gap and the “single occupation” gap all at once, then closing it is one of the most important pieces of structural work this country can take on in the next decade. It is also one of the most operationally tractable. The tools exist. The talent exists. The land base exists. The market exists. The retail rails exist and are opening. What is needed is the operating model that connects them — and the patience, the discipline and the human touch points to put that operating model into the hands of the people who will run it. That is the work in front of us at Lionheart Farms, and we look forward to doing it in the company of the readers of this paper.