MAgPIE, the Model of Agricultural Production and its Impact on the Environment, is an open-source framework maintained by researchers associated with the Potsdam Institute for Climate Impact Research. It models global land systems, explicitly representing agriculture and forestry and linking economic drivers to spatially explicit biophysical constraints.

The basic modelling question is large: given demand for food, bioenergy and timber, technological change, production costs, crop yields, forest growth, land and water constraints, what patterns of production and land use can satisfy those demands at minimum cost?

That makes MAgPIE very different from a farm calculator or a statistical forecasting package. It is an optimisation model embedded in a wider modelling architecture. It is coupled to LPJmL, a grid-based vegetation model, and uses biophysical information at 0.5-degree resolution. Regional economic conditions and demands enter the model, while the optimisation determines land-use patterns, yields and production costs consistent with those conditions.

For researchers interested in land-use policy, the repository is valuable even if they never run the full model.

The first lesson is modularity. Large models become difficult to audit when assumptions are buried in a single script. MAgPIE separates model components, configuration and scenario settings. New scenarios can be defined through configuration rather than by rewriting the entire model. That sounds mundane, but it is one of the differences between code that produces one paper and infrastructure that can support a research programme.

The second lesson is reproducibility beyond code. The README does not pretend that publishing source files alone makes a model reproducible. It documents software requirements, installation, configuration, execution and outputs. The project provides tutorials, versioned documentation, a citation file and guidance for modifying the model. It also explicitly encourages model-based publications and consulting work to make source code available. The model is licensed under GNU AGPL v3 or later, with an additional MAgPIE licence exception documented in the repository.

The third lesson concerns data provenance. MAgPIE’s contribution guidance says model dependencies should be publicly available where possible and warns against input data that cannot be shared. Its data-processing workflow is designed around reproducibility rather than treating data preparation as an invisible stage before the “real” modelling begins.

There are important barriers. This is not a lightweight Python package that a new user can install in five minutes. The main model is written in GAMS, uses R around the workflow, and currently requires a GAMS licence with the CONOPT solver. The developers recommend substantial computing resources, including at least 16 GB of memory. For many researchers, especially those wanting a transparent teaching model or a small national application, that is a significant entry cost.

Its scale also matters. A global land-system optimisation model answers different questions from a spatial microsimulation of farms or a parcel-level land allocation model. MAgPIE can explore internally consistent global transitions and competition among food, forestry and bioenergy demands. It should not be read as predicting which individual farmer will change enterprise or which exact parcel will convert.

That distinction is itself useful. Good modelling starts by matching resolution to the decision being studied.

The repository is therefore worth exploring less as a piece of code to copy than as an example of research architecture: modular scenarios, explicit assumptions, documented dependencies, reproducible data preparation and outputs that can be traced back to the model. Those practices travel well across modelling traditions.

That shift is already visible in education. Generative AI has moved from novelty to ordinary infrastructure remarkably quickly. Students can use it to explain a difficult concept, critique an essay, generate practice questions or simply produce an answer. Those uses are not equivalent, yet public debate often compresses them into a single question: is AI good or bad for learning?

Recent developments suggest a more useful phase of research is beginning.

On 8 September 2026, OpenAI announced a $5 million programme for independent research on how generative AI affects people aged 13 to 17. The proposed research areas include emotional development, relationships, patterns of use, safeguards, AI literacy and differences across cultural and socioeconomic settings. The programme explicitly invites experimental, observational, qualitative and mixed-method work.

The announcement is notable not because company-funded research can settle the question. It cannot. OpenAI has an obvious institutional interest in how evidence about its technology develops, which makes disclosure, independence, publication and replication particularly important. The programme itself acknowledges this by requiring an independence and conflicts disclosure, including independence and credibility among its review criteria, and strongly encouraging researchers to make their findings publicly available.

What matters is the research question it reflects. “AI use” is too broad an exposure to be scientifically satisfying.

Imagine two students who each spend thirty minutes with the same model. One asks it to solve a problem and copies the answer. The other attempts the problem first, asks the model to identify weaknesses, challenges its explanation and revises the work. Recording both as thirty minutes of AI use would hide the mechanism that actually interests us.

The same difficulty appears in emerging evidence on educational outcomes. A 2026 meta-analysis in Humanities and Social Sciences Communications reported generally positive pooled effects of generative-AI-supported approaches on outcomes including academic achievement, higher-order thinking and writing. That is useful evidence, but meta-analysis does not make heterogeneity disappear. Effects can depend on the intervention, subject, learner, comparison group, study quality and the way the AI is incorporated into teaching.

There is also a measurement problem created by the speed of the technology itself. A study designed around one generation of models can be published into a world of more capable systems. Interfaces, safeguards and user behaviour change too. Evidence can therefore age unusually quickly even when the underlying study was rigorous.

For universities and schools, this argues against two easy responses. One is prohibition based on the assumption that every use substitutes for thinking. The other is adoption based on the assumption that access automatically improves learning. Both jump ahead of the evidence.

A better approach is to specify the outcome and the mechanism. Are we interested in exam performance, retention six months later, writing quality, confidence, critical reasoning, time saved, social interaction or dependence on external assistance? Does AI provide feedback after an attempt, or replace the attempt? Are gains concentrated among students who already know how to evaluate an answer? Do effects persist when the tool is removed?

Those are less dramatic questions than asking whether AI will transform education. They are also more answerable.

The broader lesson extends beyond classrooms. As AI becomes embedded in research, public services and professional work, capability benchmarks will tell us only part of what we need to know. A system can perform impressively in isolation while producing ambiguous effects once humans reorganise their behaviour around it.

The next frontier of AI research is therefore partly social science: not merely measuring the machine, but measuring the human-machine system that forms around it.

That simple mismatch is one of the harder problems in agricultural environmental policy. Governments usually pay individual landholders, because contracts, eligibility and accountability need identifiable participants. Yet many of the outcomes being purchased are partly landscape outcomes.

Ireland’s Agri-Climate Rural Environment Scheme, or ACRES, is an interesting attempt to work across that boundary. ACRES is Ireland’s agri-environment climate scheme under its CAP Strategic Plan. The Department of Agriculture describes it as a €1.5 billion programme intended to support up to 50,000 farm families while addressing biodiversity, climate, air and water objectives. By Budget 2026, the government reported almost 54,000 participating farmers and allocated €280 million to the scheme for that year.

The important design feature is not simply its size. ACRES combines actions on individual farms with a Co-operation approach in selected areas. In February 2026, the Department opened a second application window for what it calls Landscape Actions. These are intended to address environmental problems whose scale is larger than a single parcel. The examples given by the Department include invasive species, scrub encroachment in species-rich grassland, threatened species and water-quality protection.

This matters because environmental effectiveness and administrative convenience often point to different spatial scales.

Consider water quality. A farmer can change nutrient management on one holding, but the condition of a river reflects pressures accumulated across a catchment. The same logic applies to habitat connectivity. Protecting one good patch of habitat may achieve less if neighbouring patches disappear. Conversely, coordinated actions across several holdings can create an ecological result that no participant could deliver alone.

There is also an economic problem. If environmental benefits spill beyond the farm undertaking the action, the individual farmer may have little private reason to supply the socially desirable amount. Agri-environment payments are partly a response to that gap. Landscape coordination adds another layer: it tries to align multiple private decisions with an outcome that depends on their combination.

ACRES does not eliminate the difficulty. Coordination has costs. Someone has to identify the relevant landscape, decide which actions fit it, work with landholders and judge whether the intervention is succeeding. Results-based payments also raise difficult questions about how much of an observed ecological outcome can reasonably be attributed to management rather than weather, soils or other factors outside a farmer’s control.

There is a second caution. “Landscape-scale” should not become a synonym for centrally prescribing one preferred land use across a large area. Agricultural landscapes are heterogeneous. The same intervention can have different costs and environmental returns depending on soils, existing production, farm structure and what is nearby. Better coordination therefore requires better spatial evidence, not simply larger planning units.

That is what makes the Irish experiment worth watching. The interesting question is not whether farm-level or landscape-level policy is superior. Both are needed. The more useful question is which environmental problem belongs at which scale.

Agricultural policy has traditionally been good at counting hectares, animals, payments and participants. The next step is harder: recognising that the environmental value of a hectare can depend on what happens beyond its boundary.

A personal website makes it easy to write as though ideas arrive from nowhere. They do not. The questions I ask, the models I use and the policy problems I encounter are shaped partly by the research environment in which I work.

I am doing my PhD within FUSION, the FUture Sustainable economy Integrated mOdelling Network at the University of Galway. The group brings together work in life-cycle assessment, integrated assessment, land-use modelling and environmental software. What I find useful about that mix is that an agricultural or climate problem does not have to remain inside one disciplinary box.

Much of my PhD also connects to FORESIGHT, a project funded by Ireland’s Department of the Environment, Climate and Communications under a services contract to provide agriculture and land-use modelling to the Climate Action Modelling Group. The project is led by FUSION with the University of Limerick and FERS Ltd.

That funding relationship is worth stating plainly. I expect to write about Irish agriculture, climate and land-use policy on this site. Readers should therefore know that I work within a government-funded modelling programme connected to those questions. The views I express here are my own and do not represent the position of the University of Galway, the Department, funders or collaborators.

One of the main modelling tools in this environment is GOBLIN, which links agriculture, forestry and other land uses within a national land-balance framework. My own interests increasingly begin at the point where a national pathway has to meet geography and heterogeneous farms. A nationally coherent change may look very different when it reaches dairy, beef or other production systems in different parts of the country.

That is the part of the research environment I value most: models are treated as tools for interrogating difficult decisions rather than as ends in themselves. A result is more useful when its assumptions are visible, when alternatives can be tested and when the limits of what the model claims are clear.

This site is much broader than that work. I will write about policy, economics, technology, Ghana, Ireland, the United States, land use, useful code and repositories, books and personal observations. But when my writing touches Irish climate or agricultural policy, this is the institutional context readers should have in mind.

What changed during my PhD was not a move from the past to the future. It was the distance between the future being modelled and the conditions contained in the data.

Estimated relationships are most defensible around the range of variation from which they were learned. As a policy experiment moves farther beyond that support, the exercise becomes increasingly dependent on extrapolation and modelling assumptions. A regression can still produce a number, but the fact that it can calculate an answer does not mean the historical data have directly identified behaviour under those new conditions.

That problem matters in climate and land-use policy because some plausible pathways involve changes much larger than ordinary year-to-year variation. Ireland’s Land Use Review, for example, examined scenarios involving large-scale rewetting of grassland on organic soils and substantial additional forestry by 2050. Those are not small marginal adjustments around today’s land system. They are structural changes whose consequences cannot simply be read from a historical coefficient.

Source: Daniel Hanrahan / Wikimedia Commons, used under CC BY 3.0.

This is where scenario modelling becomes useful. A scenario model can start from a policy target or a set of explicit assumptions and ask what follows if those conditions hold. Backcasting takes the logic further by beginning with a future objective and working backwards to the changes that would have to occur between that future and the present. The important point is that these exercises are not predictions. They are structured ways of exploring futures that historical data alone cannot reveal.

The two approaches therefore need each other. Scenario models are only as credible as the assumptions they contain, and behavioural assumptions should be disciplined wherever possible by evidence from the observed world. Econometric evidence helps tell us how people, firms or farms have responded to incentives and constraints we have actually seen. Scenario models then make explicit what must be assumed when the policy question moves outside that experience.

Climate modelling has long faced a similar distinction between top-down economic approaches and bottom-up technology or sector models. The OECD notes that the two approaches answer different questions: top-down models emphasise economy-wide coherence, while bottom-up models offer more technological detail. Neither view is complete on its own.

So I do not think of my work as moving away from econometrics. It feels more like moving from a single window to a room with several. Econometrics shows what observed variation can teach us. Scenario models ask what different futures would require. Spatial models ask where change would fall, and microsimulation asks how one policy can land differently across thousands of heterogeneous farms.

The skill is not choosing one window. It is knowing what each can show, and recognising where its view ends.