Toolkit user guide

The Toolkit can support earlier recognition of risks, stronger stakeholder engagement, better coordination between roles and clearer records for review and learning.

It also provides a structure for developing shared understanding of responsible agricultural GenAI decisions. These are intended benefits, not guaranteed outcomes or a formal training qualification.

Start with the task or concern your team needs to address. This may involve data suitability, risk, stakeholder participation, accountability or legal and regulatory questions.

On Activities, use the previews to identify relevant resources. Each preview explains the activity’s purpose, intended users, lifecycle position and expected benefit.

Review who should contribute, then access the full activity materials for the questions, instructions and supporting examples. Revisit relevant resources as your project and its operating context change.

Materials and completion resources support team discussions and records of decisions. Available formats may include:

• Download a workbook for using offline with editable Word templates for team completion.
• Access the Miro board template for shared discussions and contributions.

Use the activity previews to identify relevant participants. Different roles bring complementary perspectives.

Role	Relevant perspective
Chief technology officers and technical leaders	Governance, reliability, deployment risk and organisational readiness.
AI ethics officers and boards	Ethical and AI safety considerations across the project.
Data scientists and machine learning engineers	Dataset provenance, bias, technical choices, documentation and traceability.
Product managers	User needs, local agricultural realities and operational constraints.
Research organisations and think tanks	Transparency, reproducibility and trustworthy research or policy analysis.
Business development teams, implementation partners and demonstration leads	Explaining data sources, recommendations and safeguards during pilots and rollouts.

Include stakeholders whose knowledge and experience are relevant to the activity. Local perspectives can help teams recognise assumptions, exclusions and practical constraints.

Activities are organised into three phases and eight subphases. This structure helps you locate resources; it is not a requirement to complete them in a single linear exercise.

Design

Set the foundations for a responsible, valuable and feasible AI solution.

• Problem definition: resources relevant to the problem, intended use, users and project foundations.
• Data collection: resources relevant to acquiring data, understanding its quality and informing a data plan.

Development

Build, test and validate the system using reliable data and appropriate methods.

• Data pre-processing: resources relevant to data preparation and quality.
• Model development: resources relevant to model choices, testing and refinement.
• Model evaluation: resources relevant to assessing the model against defined requirements.

Deployment

Release, operate and improve the system, and plan for its end-of-life.

• Model deployment: resources relevant to readiness for real-world use.
• Monitor & maintain: resources relevant to performance, safety, feedback and ongoing impacts.
• Model retirement: resources relevant to the model’s end-of-life.

Continuous feedback and improvement

Learning from real-world use and stakeholder feedback informs work across the lifecycle. It is not a ninth subphase.

The checklists offer complementary ways to examine a project:

• Risk identification and mitigation: potential harms and safeguards.
• Participatory co-creation, feedback and model refinement: stakeholder involvement and learning.
• Legal and regulatory compliance: relevant requirements and issues for review.

They support discussion within your wider project processes. Completing them does not, by itself, demonstrate legal compliance or guarantee responsible outcomes.

Download the Workbook that contains the three checklists as an editable Word document

For the Toolkit’s purpose and relationship with FAIR, read the Overview.

The Toolkit supports clearer records of decisions, supporting evidence, assumptions, limitations, trade-offs and responsibilities.

These records underpin traceability, documentation and accountability. Detailed activity materials provide the associated prompts and instructions for team working.

The following fictional examples illustrate different agricultural uses of GenAI. They help you identify a relevant context; the detailed applications belong with the full activity materials.

AgriConnect: farmer and extension advice

An agricultural platform introduces GenAI recommendations and summaries for farmers and extension officers. The example highlights risks involving unsuitable advice, data representativeness, consent and responsibility.

AgroThrive: policy and evidence synthesis

A team introduces GenAI to synthesise agricultural research and policy materials. The example highlights bias, missing perspectives, uncertainty and misleading impressions of consensus.

NGBT: crop education and outreach

A project explores GenAI outreach about a new barley crop variant. The example highlights balanced information, influence on adoption decisions, local practices and farmer agency.

Explore the activity previews to identify resources for your task, project roles and lifecycle phase.