AI Agents and Task Automation in K-12 Education

This hands-on workshop explores how AI agents can support learning, pedagogical work and school administration. Participants will move beyond isolated prompts to understand how agents can follow multistep processes, consult defined knowledge sources, use authorised tools, generate outputs and request human approval when necessary.

The central feature of the workshop will be the live development of several working agents for relevant school tasks. Participants will observe and contribute to the complete process: identifying the need, mapping the workflow, configuring the agent, connecting information sources, defining safeguards, testing performance and improving the design.

The emphasis will be on purposeful, bounded automation. Agents will support educators and school teams without replacing professional judgement, meaningful human interaction or accountability for consequential decisions.

Objectives and Contents

• Distinguish between conversational assistants, automated workflows and AI agents.
• Explore current capabilities in multimodal, research-based and agentic AI.
• Identify high-value applications across learning, teaching and administration.
• Analyse a process in terms of inputs, actions, decision points, outputs and approval stages.
• Configure an agent's purpose, role, instructions, knowledge, tools and boundaries.
• Determine what an agent may do independently and when it must pause or escalate.
• Build agents that work with documents, structured data and approved information sources.
• Test agents for accuracy, reliability, privacy, bias and unexpected behaviour.
• Establish monitoring, documentation, accountability and human-oversight processes.
• Consider how agentic systems may change school roles, workflows and organisational structures.

‍

Live Agent-development Laboratories

The workshop will combine concise presentations, demonstrations, collaborative analysis and an extended design studio. Participants will bring or select an existing lesson and develop a classroom-ready differentiated learning experience through the following process:

• Learning agent: An inquiry coach, project companion or practice assistant that questions, scaffolds and provides feedback while preserving student reasoning, productive struggle and intellectual ownership.
• Pedagogical agent: An assistant supporting lesson and resource adaptation, differentiated pathways, formative-evidence analysis, misconception identification or feedback preparation. Its outputs remain subject to educator review.
• Administrative agent: A bounded agent supporting an appropriate process such as policy retrieval, document classification, meeting follow-up, frequently asked questions, communication drafting or the organisation of non-sensitive information.

Each agent will be developed live using an accessible no-code or low-code environment and synthetic or non-sensitive data. Participants will contribute use cases, instructions, knowledge sources, test scenarios and improvement proposals. Depending on the available platform, participants may also develop their own small-group prototypes.

Development and testing process

• Define the user need and intended outcome.
• Map the existing process and identify unnecessary steps.
• Specify authorised data, tools and actions.
• Write the agent's operating instructions and decision rules.
• Establish prohibited actions and human-approval points.
• Test normal, ambiguous, incomplete and adversarial requests.
• Examine failure modes, inaccurate outputs and unintended consequences.
• Refine the configuration and determine whether the agent is ready for a limited pilot.

Singapore Context and Responsible Implementation

Examples will show how AI-enabled capabilities within the Student Learning Space, including Learning Assistant, Authoring Copilot, Data Assistant and feedback assistants, can contribute to wider learning and pedagogical workflows. These will be considered alongside other school-approved agent-building, productivity and automation platforms.

Ethical and governance discussions will address student and staff privacy, purpose limitation, data minimisation, external providers, access controls, retention, bias, transparency, unauthorised actions and excessive autonomy. Designs will reflect Singapore's emphasis on bounded autonomy, meaningful human control and clear accountability for agentic AI.

Participants will leave with several demonstrated agent models, at least one prototype or detailed configuration, reusable development templates, a testing protocol and a practical framework for identifying and implementing responsible automation opportunities in their schools.

Who Should Attend

• Preschool, Primary and Secondary School Educators
• Executive and Administrative Staff