DevLearn 2026: AI in L&D and Workplace Performance

DevLearn’s biggest contradiction: Create training faster, or stop creating so much of it?

Look through the DevLearn 2026 agenda, and you’ll find sessions on automated authoring, AI-generated learning materials, and faster instructional design. But you’ll also find More Training Won’t Fix Performance! (403) and Stop Adding Courses (412).

There’s real evidence that AI can reduce instructional preparation time.

In an October 2026 randomized controlled trial, researchers evaluated an AI lesson-planning assistant involving 464 teachers at 108 English primary schools. Teachers assigned access to the tool spent approximately 24% less time preparing lessons and resources, without a detectable reduction in resource quality.

But the study also found something interesting. Teachers used the tool for only around 7% of lessons in an average week, and usage declined during the trial. Many preferred adapting existing materials to creating something from scratch.

The study concerned schoolteachers rather than corporate L&D teams, so we shouldn’t assume identical results. Still, it demonstrates both the potential benefit of AI-assisted production and the importance of how that technology fits into people’s existing work.

Now consider a customer support team struggling to meet its service targets. Management requests additional training on handling calls efficiently. But the company’s software requires agents to navigate multiple screens, its procedures involve unnecessary approvals, and its performance targets don’t account for how long those steps take.

Better training might help employees use the system more proficiently. It can’t eliminate required approvals, redesign the software, or make unrealistic performance targets achievable. Those problems require decisions that L&D may have little authority to make.

The risk is that lower production costs encourage organizations to substitute instructional content for changes that would make the underlying work easier.

AI’s bigger opportunity is redesigning how learning works

Generating a lesson plan or training video is becoming a relatively straightforward application of AI. Frontier models can also write code, interact with software, analyze data, and build tools that previously required substantial development expertise.

That’s a different kind of opportunity for L&D.

At DevLearn, Vibe Coding with APIs (801) explores building custom applications connected to external services. The Rise of the Learning Engineer (613) examines how learning professionals can combine instructional design with software engineering.

Consider a customer support team that needs practice handling unusual refund requests.

Instead of assigning another presentation about the refund policy, an instructional designer could use AI-assisted coding to prototype an interactive practice tool. Employees could work through simulated requests, make decisions, and receive feedback based on approved company rules.

The designer would still need to specify the rules, test the decision paths, and verify the feedback. The system would also need maintenance whenever the policy changed. But the deliverable would give employees opportunities to apply the procedure, rather than simply read about it.

Video could support that experience when showing a situation would be useful. Visla’s video creation API, for example, supports generating video from scripts and other inputs as part of a larger application workflow.

These possibilities don’t eliminate instructional design. They give designers more ways to address a problem, including approaches that weren’t previously practical for teams without dedicated software developers.

Easier authoring creates a difficult question: Who owns the result?

Once more people can create instructional materials, organizations have to decide who is responsible for reviewing and maintaining them.

DevLearn’s Clean Up Your LMMess Using Learning Governance (204) provides a concrete example. Amanda Archer of Advocate Health describes how learning management systems can accumulate overlapping courses when subject matter experts request new content for their own audiences, even when suitable material already exists.

The session will examine a governance model implemented at a large healthcare system.

Consider what happens when two departments publish slightly different versions of a compliance procedure. Both may have been accurate when created. After the policy changes, one department updates its course while the other doesn’t.

An employee following the outdated instructions may have no way to identify which version is authoritative.

A practical governance process could require teams to check existing resources before approving new ones, identify a responsible content owner, and establish who can authorize revisions. Those decisions can prevent duplication and conflicting instructions before they reach employees.

AI might help identify overlapping material, but it can’t independently resolve disagreements about which department has authority over a policy.

There’s a related question about choosing the appropriate production format.

Suppose an L&D team needs to demonstrate a sensitive conversation between a manager and an employee. Carefully directed scenes and dialogue might justify the additional effort required by Visla’s AI Director Mode. But that production effort would serve a specific purpose: accurately depicting a situation learners need to understand or practice.

The team would still need to decide whether watching the conversation is sufficient or whether learners need to respond and receive feedback themselves.

We’ve explored this distinction in our editorial, When AI Can Make Anything, Taste Is Only the Beginning. AI can contribute creative ideas and evaluate alternatives, but people need enough expertise to judge its recommendations and know when additional production effort is worthwhile.

Measuring learning transfer is harder than counting completions

If L&D teams are going to claim that their interventions improve performance, they need evidence that people can apply what they’ve learned.

That sounds straightforward until you try measuring it.

DevLearn’s Mastering Skill Assessment (210) addresses the limitations of conventional knowledge checks. Meanwhile, Not Ideal But Oh, So Real: After-the-Fact Measurement Hacks (804) tackles evaluation when teams have incomplete data and no clear initial targets.

Why is learning transfer so difficult to measure?

Learning transfer means applying knowledge or skills acquired during training to actual work. Passing an assessment doesn’t necessarily establish that someone can perform the task under normal workplace conditions.

A realistic simulation can provide stronger evidence of capability than a multiple-choice quiz. But employees might behave differently when handling real customers, facing time pressure, or dealing with unfamiliar exceptions.

Researchers have struggled with measuring training transfer, too. A 2026 systematic scoping review of 45 studies found considerable variation in the instruments researchers used. Some measured whether employees actually applied their new skills at work, while others assessed conditions thought to influence transfer, such as motivation, supervisor support, or opportunities to practice.

Those conditions can help researchers investigate why training succeeds or fails, but measuring them doesn’t establish that employees have changed their behavior. A survey showing strong managerial support for training, for example, can’t tell us whether employees subsequently applied a new procedure correctly.

Even when workplace performance improves after training, establishing causation is difficult. A reduction in customer complaints might also reflect changes to staffing, software, or the kinds of issues customers are reporting.

These limitations don’t make evaluation pointless. They affect what conclusions the evidence can support.

L&D teams can assess whether learners perform a skill, examine changes in workplace behavior, and investigate plausible explanations for those changes. What they shouldn’t do is present a persuasive executive report as stronger causal evidence than the underlying data allows.

Customer education shows why access to useful information matters

DevLearn 2026 includes a dedicated Customer Education track developed with the Customer Education Management Association (CEdMA).

One session, Customer Education Is Distinctive (108), examines what changes when learners are customers rather than employees. Customers generally choose whether to participate, and they may encounter instructional materials across several disconnected platforms.

There’s useful evidence about their behavior, although we should distinguish customer support research from formal customer education.

In a Gartner survey of 5,728 customers, 73% reported using self-service at some point during a support interaction. Yet only 14% of service issues were fully resolved through self-service. Among unsuccessful self-service experiences, difficulty finding relevant content was a common obstacle.

Those findings don’t establish that customers prefer video tutorials over courses, or that any particular instructional format would improve resolution. They do show why simply publishing more support content may be insufficient.

A software company might already have an excellent walkthrough explaining a particular feature. If customers can’t find it when they’re stuck, its production quality won’t help them.

For procedures that change frequently, the team might also prioritize material that’s straightforward to revise. Visla’s Screen Recorder and Screen Step Recorder provide ways to document software interactions and turn them into instructional walkthroughs.

The harder decisions concern where that material appears, whether customers can locate it, and whether it addresses their actual problems.

L&D’s expanding responsibilities require more than new job titles

Several DevLearn sessions describe learning professionals taking on responsibilities beyond traditional instructional production.

The Strategic Learning Architect (710) proposes a more results-oriented role for L&D, while Purpose-Paced Leadership (803) addresses the pressure created by AI adoption and rising expectations amid limited capacity.

But greater responsibility doesn’t necessarily mean greater authority.

Suppose an L&D team investigates poor customer service performance and concludes that the company’s software and operational procedures are responsible for much of the problem. Management refuses to replace the software, won’t simplify the procedures, and continues demanding better performance through training.

The learning team can present its findings, propose alternatives, and explain which outcomes training could reasonably affect. It can’t force management to address those underlying problems.

AI might help analyze the workflow or prototype improvements, but it can’t guarantee that a technically feasible solution will be adopted. The decision to change staffing, replace a vendor, or revise unrealistic targets still belongs to the organization.

This also complicates calls for L&D professionals to become strategic partners. If they’re expected to improve business performance, they need some influence over the conditions affecting that performance. Otherwise, organizations risk holding them responsible for outcomes they don’t control.

The Broken Career Ladder adds another concern. If AI automates entry-level assignments, organizations may remove the work through which junior professionals previously developed expertise.

The session proposes deliberate practice, coaching, and progressively greater responsibility as possible responses. Those approaches require experienced staff and time, precisely the resources that overloaded teams may lack.

Visla will be at DevLearn 2026, and we’ll be looking closely at how practitioners are navigating these problems. We’re especially interested in where AI is changing the work L&D teams do, how organizations decide which instructional materials deserve investment, and what evidence teams are using to evaluate their results.

We’re also hoping to hear about the limits of what teams can accomplish. When L&D professionals identify problems outside their control, how much influence do they actually have over the decisions needed to fix them? That’s one of the questions we’ll be bringing to Las Vegas.