Agentic production for multimedia learning

by Serge de Beer, LearningTour, the Netherlands.

From learning objectives to personalised learning experiences, powered by AI agents

The way we create educational media is changing rapidly. Where production once required significant time, people, and budget, a new approach is emerging: agentic production. In this approach, AI agents actively contribute to the entire process, from the initial analysis of the learning objective to the final delivery to the student.

This does not mean AI replaces the creator. It means we can create things that were previously not feasible. Think of adaptive instructional videos, personalised podcasts, or even interactive VR experiences tailored to individual learners.

From learning objective to production

Everything starts with a clear learning objective. What should the student be able to do? In an agentic approach, this objective is not only defined but immediately translated into observable behaviors and criteria.

AI agents support this process by:

• analysing and structuring learning objectives
• incorporating prior knowledge of the target audience
• suggesting appropriate media formats

This results in a production blueprint that guides all subsequent steps.

AgentTrainer: the core of the process

At the heart of this approach is AgentTrainer, both a tool and a design framework.

With AgentTrainer, you define in detail:

• what needs to be learned
• what “good” performance looks like
• how this performance is assessed

This is done through rubrics. These rubrics are used not only for assessing students, but also for training AI agents. Instead of relying on general-purpose AI, agents are trained using extreme fine-tuning. They learn deeply within a specific domain and task, based on clearly defined criteria.

For example:

• a script agent learns how to explain complex concepts for a specific audience
• a video agent learns how to structure effective instructional content
• a VR agent learns which interactions support understanding

The result is output that aligns closely with both the learning objective and the learner. 4C-ID as a training principle for agents and the model plays an important role in how agents are trained.

Its influence is visible in:

• training agents with realistic, whole-task scenarios
• breaking down complex skills into manageable parts
• providing targeted support during training
• gradually increasing complexity

In this sense, agents follow a structured learning process themselves. This leads to more consistent, reliable, and context-aware output.

From production to personalization

Once the design is in place, production begins. Multiple agents collaborate:

• a script agent generates the content
• a media agent creates visuals, audio, or interaction
• a didactics agent ensures alignment with learning objectives
• a review agent evaluates output against the rubric

Because all agents operate within the same defined framework, quality remains consistent.

The real impact becomes visible in the next step: personalisation.

The same base content can be adapted for individual learners:

• level of explanation
• pace
• type of examples
• format (video, audio, interactive)

Each learner receives a version that better fits their needs, without requiring a full redesign.

Practical value for instructional designers

For instructional designers, this approach offers both speed and flexibility.

When learning objectives and rubrics are well defined, you can:

• generate multiple variations quickly
• adapt content without starting from scratch
• experiment with different formats

This lowers the barrier to innovation and enables more iterative design processes.

Not replacement, but amplification

AI agents do not replace the creator. They amplify them.

The role shifts:

• from creator to director
• from executor to designer
• from producing to orchestrating

The quality of the outcome depends on how well the process is designed and how effectively agents are trained.

What this makes possible

Agentic production enables new forms of learning:

• hyper-personalised multimedia
• adaptive learning experiences
• rapid updates of educational content
• scalable yet individualised education

The combination of clear learning objectives, strong rubrics, and focused agent training makes this achievable.

Workshop

During Media & Learning 2026: Co-creating the future of learning, the workshop Agentic production for multimedia learning will take place in the Hacker room.

The workshop is led by Serge de Beer, who has over thirty years of experience in educational multimedia and has been applying AI to improve learning processes since 2017. In this session, you will work hands-on. You will set up your first agents, train them using AgentTrainer, and experience how to move from learning objectives to working, personalized multimedia.