AI has split work into a before and an after. The way companies develop people is largely the same as it has been, but what technology makes possible is not. That raises a question most L&D teams have not fully faced: should the LMS still sit at the center of L&D?
We put that question to Karen Kocher in a live conversation moderated by Beth Summers, VP of Sales at Growthspace, and joined by Growthspace CEO Omer Glass. Karen is the founder of Kocher Strategies, an AI workforce transformation consultancy. She previously led future of work, AI transformation, and workforce acceleration at a global technology company, and she served as Chief Learning Officer at Cigna. Here is what stood out.
Rethinking the LMS and the Learning Technology Stack
The LMS has a job to do, but that job is shrinking
Karen did not say the LMS is dead. She said its job to be done is largely going away. What an LMS does well is track required learning in a way that regulators and governments accept. Other technologies can do that too, which could make the LMS unnecessary for that purpose.
“The job to be done is largely going away. Now, it’s not wholly going away.”
SCORM can become a barrier to AI enabled learning
Karen also named a deeper problem. LMS platforms package content in SCORM, and that format is tight like a drum. It works well for compliance tracking, but it becomes a barrier when you want AI enabled learning, because your content is locked inside a wrapper built for a different era.
Too many tools, too much noise
Add the LXPs and the many content creation tools most companies have collected, and the learning technology ecosystem becomes vast. Karen’s view is that it likely needs to shrink to far fewer components.
The economics of replacing software have shifted. Omer pointed to a founder who previously led a company with more than a billion dollars in revenue and now retires one software product every week. The most recent was a category leading sales platform. His engineer rebuilt the functionality in six days. A few years ago, replacing a core system meant a multi year project, so most companies never questioned whether they should.
Now the cost of staying put is visible, and it deserves the same scrutiny as the cost of change. That does not mean every tool should go. It means every tool should face three questions: What job is it doing? Does it do that job the way we want it done? And does it confine us to the assumptions it was built on? The third question matters most. A tool does more than perform a task. It shapes the process, the data, and what people believe is possible. In L&D, that is where an LMS, a content library, or a stack of authoring tools can quietly set the ceiling for the whole function.
Start With the Work, Not the Infrastructure
Be in a hurry to skill people, not to rebuild everything
One of the most useful points in the session was a caution. Karen said she would be in a hurry to skill people, but not in a hurry to rebuild the entire learning infrastructure.
Her advice is to start with the jobs to be done and the workflows around them. Ask what AI should do first, what a human is uniquely qualified to do, and what expertise your company wants to be known for. It is hard to decide how people will learn when you are not sure which expertise differentiates you.
Run experiments and prove out what works
From there, run experiments. Build or buy a tool, test whether it gets you where you are trying to go, and prove out what works. Once you have a few tools you trust, you can figure out the broader infrastructure and decide what to do with the historical content.
“Nothing’s worse than trying to unravel the historical, thinking that you have to do that before you make progress, because nobody’s going to wait for you to do that.”
Skills, Not Course Completion
Skilling is different from learning objectives
Karen made a distinction that is easy to miss. For years, L&D worked toward broad learning objectives. Very few people could say what skills a person actually needed to be good at a job. That means many teams were helping people develop, but were not really skilling them.
Once you are clear on the skills, you get agility. If the work changes slightly, you know what you can build on and what takes more.
What to measure instead
Course completion tells you someone finished a course. It does not tell you they can do the work. Karen’s answer is simple: the best test of whether someone is skilled is whether the work got done, and done well.
The person best placed to judge that is whoever knows what good looks like. That could be a manager, a project leader, the individual, an AI, an expert, or a mechanism that simulates that expert. L&D does not have to be the one to declare someone skilled.
Expertise Still Matters
Humans are hard to beat as teachers
Omer pointed out that a trainer led experience is usually great, whether the teacher is a manager, a mentor, or a trainer. Watching a video or reading an article is usually not.
“The promise of AI is not to replace the human. It is to make the non human more human like.”
Omer Glass, CEO and Co Founder, Growthspace
The apprenticeship model still works best
Karen agreed that the best way to build expertise is the old apprenticeship model. If you want to build a skill, find someone who already has it. The learning function has always struggled to stay ahead of what people need to learn, because the people with the skill use and sharpen it every day. That is why Karen argues for an infrastructure built around expertise.
Deciding what stays human
The practical question is balance. For any expert led program, decide which part the expert must do as a human and which part AI can handle. Part of that decision is cost, and part of it is protecting the time of experts who are also trying to deliver value to the company.
What It Looks Like to Build From a Blank Sheet
Learning inside the flow of work
Karen described a concrete case. A team wanted technical support engineers to build skill every day without ever sitting through a formal program. No course, no two minute video. The goal was for AI to work inside their real cases and build knowledge in real time.
An AI expert modeled on a sales leader
A second example came from a sales organization, where an AI expert was built to emulate the head of sales. It asked salespeople questions, offered coaching, and acted like mentorship. Karen described it like being stopped in an elevator by a leader who asks what you are about to walk into.
“You would have left that elevator being like, ‘Man, I know exactly what to do.’ That’s the way the agent works.”
The key difference is that this works in an anticipatory way. A prompting model assumes you know enough to ask the right question. An expert that works alongside you does not.
Reaching thousands of people at once
This approach also reaches far more people than an SME ever could. Karen noted that the long held hope was for subject matter experts to create all the content themselves, but they have other work to do. Capturing their knowledge in an AI expert reaches tens of thousands of salespeople at once.
Karen was clear that self paced learning still has value. But in a world that changes this quickly, where expertise is a differentiator, people need other ways to get it.
L&D Cannot Wait to Be Told What the Work Is
The business is still figuring it out
Karen described a real tension. Most CEOs know what success looks like in business outcomes. Few know the best way to get there. If the work itself is being redesigned, L&D cannot wait to be briefed, as it historically did.
Lean in with experience, not just reading
Her advice is to lean in. Help the business redesign workflows and identify skills, because L&D has real strengths there.
She added a caution about credibility. If you are going to challenge how things are done, speak from experience. Redesign the workflow of one job, learn from it, and bring that story to the table.
“Nobody wants the questioner to be simply somebody who’s read the books.”
Practical Lessons for Moving Faster
Call it an experiment, not a pilot
A practical tip came up when Beth Summers raised the stigma around pilots, and Karen built on it. At her previous company, the word pilot carried a hidden meaning: the first of many rollouts. People hesitated because a failed pilot looked like a bad decision. Calling it an experiment changed the mindset. Trying something that does not work simply tells you what is possible, and you move on to the next idea.
Use as few labels as possible
The same applies to the language you use for the work itself. Karen’s advice is to use as few labels as possible, and to use words the business already knows. Skills, competencies, and workflow redesign can sit under one umbrella if you anchor it on the outcome you want.
Reduce handoffs and focus on the problem, not the tools
Karen closed with a point on orchestration. In most corporate workflows, many humans get involved from start to finish, and every handoff is a coordination tax. People forget, do it wrong, or do it late. The ideal is a workflow with as few handoffs as possible, owned by one human or by AI. Some workflows can be completely autonomous.
Where AI can take on more of the work, the human moves up to work that is uniquely human.
“It’s not about the tools. It’s about the answer we’re after and how to get there as efficiently as we can.”
What to Take From This
Define the job to be done before you pick technology. Redesign one workflow and learn from it. Measure whether the work gets done, not whether the course was finished. Find your experts and decide where they should stay human. Run experiments, and move quickly.
Omer put it simply: start with AI native solutions, think about the job to be done, and be bold.
About Growthspace: Growthspace builds ExpertX, an agentic L&D platform that turns an organization’s knowledge into AI led training, practice, coaching, and support. Teams configure the knowledge sources and delivery formats that fit their goals, from Q&A to role play. Learn more at growthspace.com/products/expertx.
Related Articles
.png)
Ready to turn insights into impact?
