iberestudios.

The future of eLearning depends on what learners can do

For years, improving an online course has meant adding more content: a clearer video, a better document, an expert lecture and a quiz that checks how much the learner remembers. Generative AI makes it easier and cheaper to produce all of these. But a larger library does not answer a more demanding question: what can the student actually do when faced with a problem for which nobody has provided the next step?

Michael W. Allen has spent decades working on the gap between explanation and learning. He contributed to PLATO, created Authorware and wrote about iterative instructional design long before generative AI arrived in educational platforms. In Rethinking eLearning: What Works. What Doesn’t. What’s Missing., published in February 2026, he returns to an old and unevenly addressed problem: how to design learning experiences that change what people do rather than simply deliver information.

Learning begins when there is a decision to make

Consider a business automation course. A conventional version begins by defining an AI agent, an API and the tools that connect systems. Then it shows a working workflow. Students answer ten questions and receive a grade. A performance-centred alternative might start at the other end: a company receives two hundred customer enquiries a week and must decide which can be answered automatically, which need review and which must be escalated to a person. Learners get sample data, privacy constraints and a budget. Their task is to design a process and explain what it will do when something goes wrong.

The theory still matters. What changes is when learners need it. If they automate every reply, the simulation might reveal a sensitive complaint sent to the wrong recipient. If they require too many manual checks, they may find the team cannot cope with the workload. Guidance on controls, exceptions and escalation now has an immediate purpose. Learning is organised around a decision that has to be revisited.

From a tutor that answers to a coach that observes

The public Treasure Trove library associated with Allen’s book brings together demonstrations of interactive learning design. Examples presented include Artha Learning’s Practice to Confidence, for practising AI prompts with feedback; Allen Interactions’ goal-based Constellation; and an ELB Learning Role Play built around management decisions. These are design references, not controlled trials demonstrating learning gains.

An AI tutor becomes more interesting when it stops solving the problem the moment a learner struggles. A coach can ask for a justification, offer a limited hint, introduce an exception or provide another test. In the automation exercise, it might ask what happens when two messages have the same subject but different privacy risks, or how a failure at an external service will be detected. The student has to change the workflow. The conversation becomes practice, and the resulting system can be tested.

A convincing explanation is not proof of competence

When one AI judges work written with another AI, a persuasive answer can easily be mistaken for a functioning solution. A more demanding assessment for an agent-building course would run the system against unseen cases, measure failures, require an explanation of key decisions and observe how the learner repairs a fault. A rubric might assess architecture, exception handling and data protection. Peer review can expose risks that the original author missed.

Educational research supports the value of guided practice, although not all simulations are alike. A systematic review of 289 studies in health-professions education, involving 18,971 trainees, reported favourable results for design features such as interactivity, repeated practice, individualisation and feedback. The researchers also reported substantial heterogeneity. This is evidence for certain design components in a particular educational context, not proof that every AI-powered simulation will work in every subject.

The challenge for schools already using AI

Educational institutions have long used projects, rubrics, mentoring and gamification. The next step may be less flashy and more demanding: design the tasks before producing the materials. What professional decision should the learner be able to make? What consequential mistakes might occur? What information is required? What evidence would distinguish a skill acquired from an exercise completed with assistance?

In marketing, a scenario might ask students to allocate a budget using incomplete evidence and respond to a sudden fall in conversions. In leadership, it could involve negotiating with an employee who has sound reasons to reject an agreement. In cybersecurity, it might mean spotting an anomaly and deciding what to isolate without shutting down the entire operation. These are proposals for applying the approach, not institutional results verified by Allen’s book.

An independent-paced AI master’s programme could start with one unit: an initial diagnostic, a functional challenge, graduated hints, peer review and a transfer test using a different case. Before extending it, the programme would need to compare what learners acquire, retain and can do without help, alongside the cost of support. The tutor’s own mistakes and premature answers should also be recorded.

A more demanding question for online education

Allen puts instructional design back at the centre just as producing content has become exceptionally easy. The number of video hours, exercises or instant chatbot replies will tell us little about the quality of the learning experience. What matters is which problems a person can solve afterwards, and how much of that ability remains when the hints disappear. AI makes this challenge more promising—and harder to conceal behind polished content.

Leer este artículo en español

Sources and further reading

This article draws on the publisher’s description, publicly presented examples and research on instructional design. It is not a claim to have read all 272 pages or conducted a full functional audit of the demonstrations.

Training application data

Óscar Fuente Castrillejo, the owner of Iberestudios, is responsible for your data. We store the information you submit to manage your beta application, place reservation or requested notification. Each form records your consent separately. To withdraw your application or request access, correction or deletion of your data, email ofuente@gmail.com.

Legal information — Spanish