Why AI makes senior instructional designers more valuable, not less

If you work in learning and development, you already know the mood has shifted. Two years ago the conversation about AI was mostly excitement. Now it is a mix of excitement, fatigue and a question a lot of people are not saying out loud: does this make me replaceable?

I want to answer that question honestly rather than reassuringly, because the honest answer is more useful. It is no, but not for the comforting reason people reach for first. AI does not protect instructional designers because the work is too creative or too human to automate. It protects senior instructional designers because the parts of the job AI is good at were never where the value lived, and the parts AI struggles with are exactly what a good senior designer spends a career building.

The mood, honestly

The L&D Global Sentiment Survey 2026 is the clearest read on where the profession actually stands, and it does not sugar-coat anything. AI is still the single most talked-about topic in L&D. But interest has peaked. Sentiment has soured compared with the earlier waves of enthusiasm. Teams are using AI more selectively than the hype suggested they would, and there is real, well-founded anxiety about what it means for jobs.

I do not think that mood is wrong, and I am not going to argue people should simply feel better about it. A profession that spent two years being told AI would transform everything is entitled to some scepticism when the transformation turns out to be messier and slower than advertised. The useful move is not to dismiss the anxiety. It is to look at what is actually happening underneath it.

What is actually shifting: from speed to impact

The clearest signal of that shift comes from the Synthesia AI in L&D Report 2026. It is a vendor report, so I am treating its numbers as directional rather than gospel, but the direction it points to matches what I see in my own work.

Today, most AI use in L&D is production work: generating voiceover, drafting quiz questions and course content, producing video, translating material into other languages. The top benefit teams report right now is faster production. That is real and it is useful. It is also, frankly, the least interesting thing AI does for learning.

What the same report finds is that teams expect the biggest future gains to come from somewhere else entirely: personalised learning, wider reach, and genuine engagement. In other words, the profession's own data shows the value of AI shifting away from "how fast can we make content" and toward "how well does this actually work for the person learning." Production speed is a floor. Learner impact is where the ceiling is, and getting there is a design problem, not a generation problem.

Adaptive learning is not a prompt, it is a project

This is where the anxiety and the evidence really part ways. There is a common assumption that AI can simply generate a personalised, adaptive course the way it generates a paragraph. It cannot, and the peer-reviewed evidence is specific about why.

A 2025 study in Computers & Education: AI found that building a genuinely adaptive course is labour-intensive and depends heavily on instructional designers working closely with subject-matter experts throughout. The study puts a number on it that stopped me when I first read it: a single adaptive course can take six or more academic semesters to build properly. Not weeks. Semesters, plural, with faculty and designers iterating together the whole way through.

That is not a limitation of the AI tools available at the time of the study. It is a description of what adaptive, effective learning actually requires: someone who understands both the subject matter and how people learn, sitting with a subject-matter expert, deciding what the adaptive branches should be, what the decision points are, what "wrong" looks like and what should happen when a learner gets there. AI can help populate that structure once it exists. It cannot design the structure, because the structure is the judgement.

The skills AI cannot commoditise

If production speed is the floor, the skills that sit above it are the ones that were always the harder, more senior part of the job:

• Judgement. Knowing when AI-generated content is subtly wrong, badly targeted, or technically accurate but pedagogically useless.
• Governance. Setting the rules for where AI is and is not appropriate to use in a given piece of learning, and being able to defend that decision.
• Accessibility. Designing so that learning actually works for the person using assistive technology, not just the person the AI-generated draft implicitly assumed.
• Subject-matter collaboration. The unglamorous work of sitting with the people who actually know the content and turning what they know into something teachable, which the Computers & Education: AI study shows is exactly where the real time goes.
• Evaluation. Knowing whether any of it worked, measured against behaviour and outcomes rather than completion rates or how quickly it was produced.

None of that is under threat from AI. All of it becomes more valuable as AI raises the baseline speed of production, because faster production means more content going out of the door, which means more places where bad judgement, poor governance or inaccessible design can do damage at scale. Someone has to be the check on that. It is not going to be the tool.

Regulated sectors make this concrete

I do most of my work in regulated environments, healthcare and care among them, and that is where this argument stops being theoretical. Two legal requirements sit underneath everything I build, and both require a skilled human, not a tool.

The first is accessibility. Public sector and much regulated-sector digital content in the UK has to meet WCAG 2.2 AA as a legal requirement, not a nice-to-have. That means someone has to actually check contrast, keyboard operability, focus order and plain language against a standard, and know what to fix when it fails. AI can draft alt text. It cannot certify that a course is accessible.

The second is oversight of AI itself. The CQC's principles for AI in health and social care are explicit that AI use in these settings requires human oversight. That is not a compliance formality. It is a recognition that AI output in a regulated, high-stakes environment has to be checked by someone who understands both the content and the risk. That someone is a skilled instructional designer working with the right subject-matter experts, not a generic reviewer and not the AI itself.

Put those two requirements together and you get a simple picture. Regulated learning cannot legally ship without the exact human skills AI does not have. That is not an argument I am making up to make senior designers feel better. It is written into gov.uk guidance and CQC's own expectations.

Where this leaves the work

I use AI in my own practice, mostly where the Synthesia data says everyone is using it: production, drafting, first passes at content that I then shape properly. I am not precious about that and I do not think anyone should be. But the work that actually changes what a clinician does when a patient is at risk, or what a care worker does when something does not look right, still comes from sitting with the person who knows the job, working out what the real decision point is, building something that lets people practise it safely, and checking, properly, that it worked and that everyone can use it.

That is the work AI speeds up but does not replace. It is also, not coincidentally, the work that was always hardest to do well, which is why it stays valuable as the easy parts get faster. Learning that changes behaviour, not just tick boxes, was never a production problem. It was always a judgement problem, and it still is.

Frequently asked questions