How AI Changes the Create or Curate Decision for L&D

AI makes content easier to produce, but L&D still needs sound judgment about what to curate, customize, create or solve without training.

Generative AI can now produce course outlines, explanatory text, quiz questions and scenario variations in a fraction of the time they once required. That new capacity creates a deceptively simple question for L&D: if content has become easier to generate, should the team build more of it, curate what already exists or direct its effort somewhere else?

The Problem Isn’t E-Learning Content. It’s Content Without Purpose

Consider two requests that reach an L&D team on the same morning.

The first comes from a regional operations leader. Managers are approving customer exceptions inconsistently, creating delays and avoidable escalations. The request is straightforward: "Can you build a short e-learning course on the correct approval process?"

Initial conversations reveal a different problem. The policy is scattered across several documents, responsibilities overlap and even experienced managers disagree about who can authorize an exception. A course could explain the process as it currently exists, but it would not resolve the ambiguity causing the inconsistent decisions. The organization needs to clarify the workflow before it trains anyone.

The second request concerns a new piece of production equipment being introduced across facilities in three countries. The manufacturer has provided a manual and several public demonstration videos. Technically, the content already exists. Yet the available material does not reflect the company's operating procedures, local safety requirements, employee roles or escalation protocols. Some employees require translated instruction, and the business needs verifiable evidence that operators can respond correctly when the equipment behaves unexpectedly.

Here, relying on existing content would leave a meaningful performance and safety gap. The learning team may need to create a focused programme combining essential instruction, demonstrations, guided practice, realistic fault scenarios and an assessment based on the organization's own procedures.

These scenarios illustrate the judgment enterprise L&D is expected to exercise. In one case, creating a course would convert a process problem into a training problem without correcting its cause. In the other, refusing to create content would leave employees without the contextualized guidance and controlled practice required to perform safely.

AI changes the speed and economics of content development, but it does not settle the create-or-curate decision. It can summarize existing material, draft a course or generate a practice scenario. It cannot, on its own, clarify an ambiguous workflow, determine whether training is the right response or decide how much organizational context and control the situation requires.

This makes a familiar L&D habit more consequential. A business request arrives, and the conversation moves quickly to modules, screens, interactions and delivery dates. With AI available, the team can reach production even faster, sometimes before anyone has established what employees need to do differently or what is preventing them from doing it.

A performance gap may arise from confusing processes, poorly designed technology, competing priorities, limited resources or unclear expectations. Training will not correct those conditions. Even when employees need new knowledge or skills, the appropriate response could be a job aid, structured practice, coaching, manager support or easier access to information during the task.

L&D teams should also question the value of using AI to recreate generic explanations already available through credible books, articles, videos, podcasts and learning platforms. Enterprise requirements, however, cannot always be met through curation. Organizational context, proprietary systems, regulatory obligations, accessibility, multilingual deployment and the need for consistent practice may make purpose-built content essential.

The challenge is to make every learning element earn its place. That requires L&D to determine what can be reused, what must be adapted, what deserves to be created and when the right response is to solve the workplace problem rather than commission another course.

Content Is Part of the Experience

Criticism of content-heavy e-learning often relies on a narrow picture of content: explanatory screens, narration, stock images, decorative interactions and a quiz at the end. That is certainly content, but it represents only one form.

Consider the alternatives commonly proposed for a new manager programme. A performance-conversation simulation requires a realistic situation, employee history, dialogue choices and credible consequences. A case discussion needs enough operational detail to support a meaningful decision. A team challenge requires instructions, constraints and evaluation criteria. Coaching depends on an observation guide or feedback model that reflects the expected behaviour.

All these elements have to be designed and developed. Their value comes from the work they enable rather than the amount of information they contain.

This functional distinction is more useful than separating content from learning design. Some content explains. Some demonstrates. Some creates a decision environment in which learners can practise, make mistakes and receive feedback. Effective programmes usually combine these functions in different proportions.

The production question should therefore change from "How quickly can we build the course?" to "What is the smallest useful intervention that can improve performance?"

A concise explanation followed by a worked example may be enough for one audience. Another may need a branching scenario and manager-led practice. Employees performing an infrequent but critical task might benefit more from a searchable job aid than from a course they completed six months earlier.

The format follows the performance requirement.

Practice Works Best When Learners Are Prepared

Practice, reflection and feedback play a central role in learning, but learners cannot always begin with an unfamiliar challenge and work out the underlying principles for themselves.

Imagine asking newly promoted managers to handle a simulated performance problem. They may be confident communicators, yet unfamiliar with the organization's documentation requirements, employee-relations policies or escalation process. If the simulation comes before this foundation is established, it may reward improvisation rather than develop reliable judgment. In a sensitive situation, the learner could rehearse language or actions that create additional risk.

Research on cognitive load and instructional guidance helps explain why. Kirschner, Sweller and Clark found that minimally guided problem-solving can impose excessive demands on novices. Learners must understand unfamiliar subject matter while simultaneously determining how to solve the problem. Explicit guidance, modelling and worked examples help them build the knowledge structures needed for more independent performance. Read the research in Educational Psychologist.

The appropriate level of support changes with expertise. An experienced manager may learn more from an ambiguous scenario with limited guidance. A first-time manager may need a model conversation, a clear explanation of the process and guided practice before facing a similar situation independently.

This suggests a progression rather than a choice between instruction and experience. Establish the essential knowledge, demonstrate how it applies, provide supported practice and then introduce situations with greater complexity. Workplace application and feedback continue the process after formal instruction ends.

Reducing unnecessary material strengthens this progression. Removing essential guidance does not.

Curation Is Design Work

The volume of available material can create the impression that organizations no longer need to develop explanations of established topics. Leadership alone offers university lectures, conference talks, books, podcasts, commercial courses and content from prominent thinkers.

Availability, however, tells us little about instructional suitability.

A compelling speaker may provide a memorable perspective without addressing the decisions managers face inside a particular organization. Two credible experts may define accountability differently. A university lecture may assume prior knowledge that employees do not possess. A forty-minute video may contain five useful minutes surrounded by material that is interesting but unrelated to the performance objective.

Curation becomes valuable when someone makes the necessary judgments. The learning designer must evaluate accuracy, relevance, level, accessibility and currency. The selected resources then need to be sequenced, contextualized and connected to an activity or workplace decision. Without that work, a resource library remains a collection rather than a learning pathway.

Research on multimedia learning reinforces the importance of coherence. Learners generally benefit when extraneous information is removed and cues direct attention to essential relationships. Moving between external sources with different terminology, assumptions and structures can increase cognitive effort without producing a clearer mental model. See the coherence and signalling principles in the Cambridge Handbook of Multimedia Learning.

For a stable, generic subject, a carefully selected external resource may be ideal. Where employees need only a small part of several longer resources, a focused explanation developed for the organization may be more efficient for both the learner and the team maintaining the programme.

"Curate before creating" is a sound discipline. The final decision still depends on instructional fit.

Enterprise Context Cannot Always Be Curated

External material is most useful when a subject is widely applicable and relatively stable. Much of enterprise learning concerns knowledge that is specific, controlled or frequently updated.

An external expert can explain the principles of giving feedback. That expert cannot tell a manager how the organization records a performance concern, when HR must be involved, which actions require approval or how local employment requirements affect the conversation.

The same limitation appears in software training. A generic tutorial may explain a platform's standard features, while employees work in a customized environment with different permissions, fields and approval paths. In safety training, a public video may introduce a hazard without reflecting the actual equipment, emergency procedures or conditions at a particular facility.

This context influences transfer. A meta-analysis of 89 studies found that workplace application depends on trainee characteristics, the design of the training intervention and the work environment. Exposure to sound ideas does not ensure that employees can apply them under the conditions in which they work. Review the meta-analysis in the Journal of Management.

Enterprise teams must also consider operational requirements. Public resources can disappear, change or move behind a paywall. They may not support the required languages, accessibility standard, assessment strategy or completion record.

OSHA, for example, states that required employee training must be provided in language and vocabulary workers can understand. Review OSHA's training compliance guidance. The Web Content Accessibility Guidelines include requirements for captions on prerecorded synchronized media and alternatives for certain time-based content. See the W3C guidance for meeting WCAG 2.2.

For a global enterprise, these are design and risk-management considerations. Controlled content gives the organization the ability to verify what was taught, adapt it for different audiences, update it when a policy changes and demonstrate that the intended employees received it.

Well Designed E Learning Supports More Than Consumption

Equating e-learning with passive information delivery understates what the medium can provide. A poorly conceived course may consist of text, images, superficial interactions and a final quiz. A well-designed programme can combine demonstrations, branching conversations, software simulations, diagnostic assessments, retrieval practice, adaptive remediation and workplace assignments.

The medium does not determine the quality of the learning. The instructional decisions made within it do.

A meta-analysis comparing web-based and classroom instruction found broadly similar results when both used the same instructional methods. Web-based instruction performed particularly well under conditions that included learner control, practice and feedback. See the Sitzmann et al. meta-analysis in Personnel Psychology.

Assessment design illustrates the difference. A quiz used only to record completion contributes little. A question that requires learners to retrieve relevant knowledge, make a judgment and receive corrective feedback serves a clear learning function. A systematic review found that retrieval practice improved learning across different subjects, education levels and assessment conditions. Read the review in Educational Psychology Review.

Authoring tools can be used to produce passive courses, but they can also support scalable practice and feedback. Their value depends on the problem being solved and the quality of the design.

AI Shifts the Development Effort

Generative AI expands what learning teams can provide. It can help create scenario variations, conversational practice, adaptive support and timely feedback. Employees can encounter a broader range of situations than a fixed scenario would normally allow.

Those opportunities come with additional design responsibilities.

An AI practice partner may invent a policy, overlook a legal concern or give inconsistent judgments about similar responses. This variability may be acceptable during low-risk brainstorming. It becomes more consequential in compliance, safety, healthcare, financial services or employee relations.

The NIST Generative AI Profile recommends considering human review, documentation, monitoring and management oversight because generative systems can produce inaccurate or misleading outputs.

A dependable AI-enabled learning experience therefore needs an approved knowledge base, credible scenarios, performance criteria, feedback boundaries, escalation rules and a testing process. Learning designers must decide when AI-generated feedback is sufficient and when a coach, manager or subject-matter expert should review the learner's performance.

AI can reduce the effort involved in drafting and variation. The work does not disappear; it moves towards validation, interaction design and governance. Instructional judgment becomes more important because the system can generate experiences at a scale that would be difficult to review manually after deployment.

Curate Create Customize or Do Not Train

A disciplined L&D response begins with the cause of the performance gap. What must employees do? What are they doing now? Are they missing knowledge or practice, or are workplace conditions preventing the required behaviour?

The answers lead to four possible decisions.

Current workplace research points in the same direction. A 2026 UK government review of AI upskilling found that stronger approaches connect learning to actual tasks, tools and organizational systems instead of treating training as an isolated event. Review the supporting workplace case studies.

This decision model also creates accountability. Curation should not become a way to avoid contextualization. Custom development should not proceed simply because a team has production capacity. AI should not be introduced because it offers a novel interaction. Each choice must explain how the intervention will help employees perform.

Make the Purpose Visible

The challenge to conventional course production is overdue. L&D teams should be prepared to question unnecessary content, reject requests that will not solve the problem and redirect effort towards practice, feedback and workplace support.

Enterprise learning still requires purpose-built content where accuracy, context, accessibility, scale or risk demands it. The standard should be whether each element helps employees understand or perform something that matters.

That standard changes the role of the learning designer. Success is no longer demonstrated by how much content was created or how quickly a course was published. It is demonstrated by the quality of the decisions behind the intervention: what was removed, what was reused, what had to be built and how those choices supported performance.

The future belongs to learning teams that can make those decisions well.