AI rollout is no longer a competitive advantage on its own. Most large organisations have already moved. They bought licences and launched pilot projects. Employees are experimenting in pockets across the business. The tools are no longer the rare part. The harder question is what happens next.
Can people use AI well enough to change the quality, speed and shape of their work? Can managers tell the difference between useful experimentation and noise? Can teams redesign workflows, rather than simply add AI on top of what they already do?
That is where many organisations are now getting stuck.
We call this gap the Learning Bottleneck. It is the distance between having AI capability available and being able to absorb it into real decisions, workflows and behaviour. Previous transformations were often constrained by capital, infrastructure or technology maturity. This one is being constrained by adaptation speed.
That matters for Heads of L&D and CHROs.
For years, the learning function has argued that capability matters, that culture matters, and that strategy only works when people can change how they work. Often, L&D has been brought in late: once the strategy is set, the technology bought, and the roadmap written.
AI makes that sequence harder to defend.
AI changes more than the tools people use. It changes how they make decisions, how they judge quality, and sometimes how they see their own expertise. It asks people to rethink what good work looks like, what judgement means, how value is created, and where human contribution still matters most.
The cost of underinvesting in the human side is already visible. In the Work AI Index, a 2026 survey of 6,000 digital workers, 87% now use AI at work and 75% say it makes them more productive. Yet only 13% say their organisation is performing significantly better as a result.1
Employees may be developing individually, but collectively, the organization has yet to become greater than the sum of its parts.
That is the uncomfortable pairing. The most learning-dependent transformation in decades is underway, and the learning function has not yet been given the mandate to lead it.
The question we hear most often from senior HR and L&D leaders is this: how does L&D move from running AI training courses to owning the organisation’s ability to learn?
Mostly, the technologists.
In a recent Zapier survey, 34% of executives said IT and engineering leadership are responsible for defining and maintaining AI skills in their organisation. Learning & Development or HR owns it in 7% of cases.1
The same survey found that 64% of organisations plan to train current employees on AI. Yet across every department, including IT itself, the share of employees receiving formal, structured training hovers around half or below.
The pressure usually starts at the top. Leadership wants to see AI progress, so they approve budgets and allocate licences. IT leads the rollout because the tools sit in its world. Then, when adoption and ROI do not appear quickly enough, the People function is asked to “fix” uptake through upskilling.
That is how L&D ends up owning the symptoms of a capability problem it was never resourced to lead. IT understands how the tools work, but capability building is a different discipline.
HR and L&D exist because subject-matter expertise and instructional expertise are different skills. Assuming one automatically creates the other is how organisations end up with a lunch-and-learn: a two-hour technical demo that nobody remembers, and even fewer people apply.
There is a second structural problem: shelf life.
78% of executives report at least one barrier to building an AI-skilled workforce. The most commonly cited barrier, at 18%, is that the pace of AI change makes training obsolete almost as soon as it is developed.
That fact should change the conversation. A course catalogue can become outdated within weeks. A learning system that helps people keep absorbing change has much longer value.
Designing systems that help people learn continuously is the work L&D has always claimed as its territory.
Try this: In your next leadership conversation, ask who made the five most important AI capability decisions this quarter.
Who decided which tools were rolled out? Who decided what level of AI fluency is expected in each role? Who decided which roles or workflows are being redesigned? Who decided where the training budget went? Who decided how success is being measured?
If L&D appears nowhere in those answers, that is the agenda for your next executive meeting.
Here is the warning we would put in front of every Head of L&D right now: do not mistake this moment for a training opportunity. If L&D's answer to AI is a bigger course library, it will confirm the support-function stereotype that produced the 7% figure in the first place. This is a moment for L&D to earn a more strategic role. The opening is for L&D to become the architecture function for organisational learning in the age of AI: the team that defines what AI capability means here, builds the systems that develop it, and measures whether it is producing business value.
The distinction between AI literacy and AI fluency carries most of the weight. AI literacy is knowing what the tools are. AI fluency is knowing how to use them to create better work: to redesign a workflow, sharpen a decision, raise the standard of an output. Awareness programmes produce literacy. Transformation requires fluency, and fluency differs by role, team, seniority, and process, which is exactly why it needs to be architected rather than broadcast.
The organisations pulling ahead treat AI fluency as something built into how people work, not taught once and hoped for. The Work AI Index calls this the human infrastructure of AI, the context-setting, judgement, and standards that make AI worth using and its central finding is blunt: it can't be bought, it has to be built.
One company worth studying, briefly, is Zapier. In October 2025 it promoted its Chief People Officer, Brandon Sammut, to the newly created role of Chief People & AI Transformation Officer, an explicit signal that AI transformation belongs to the people agenda as much as the technology agenda.
More instructive than the title is the mechanism: a four-level AI Fluency Rubric embedded into the operating system of talent, used in hiring, onboarding, and performance expectations, with the company reporting that 97% of employees use AI daily (Zapier HR-led AI transformation playbook). Sammut has described AI fluency as less a matter of prompts and models than of knowing work well enough to redesign it. The lesson generalizes well beyond one mid-sized software company: fluency became universal because it was built into how people are hired, evaluated, and expected to work, rather than taught in a classroom and hoped for afterwards.
The same research finds that the highest-performing AI users learn less from formal courses than from the work itself: they are more than twice as likely as their peers to rate AI as a valuable teacher (68% versus 28%), because every prompt is practice and every poor output is feedback. Fluency, in other words, is a by-product of guided practice on real work, precisely the system L&D is built to design.
This is also the way out of pilot purgatory. Most AI pilots stall not for technical reasons but because no one owns the behaviour change required to scale them. Scaling a pilot is, at its core, scaled behaviour change: someone has to own the system that turns isolated wins into common practice. That is the seat L&D should be asking for
Try this: Draft a one-page definition of AI fluency for your organisation, with three levels of proficiency for each major role family. Keep tools out of it; describe what people can do at each level. Bring it to your CHRO or executive team and ask one question: who owns moving the organisation up this scale?
There are fair objections to this argument. L&D leaders will meet all of them, and they deserve straight answers.
“L&D lacks the technical credibility.”
Partly true, and partly beside the point. L&D does not need to out-engineer the engineers. The capability question, “what should people be able to do differently?”, is distinct from the tooling question, “what should we deploy?” The credible model is joint ownership: IT owns the stack, L&D owns the capability system, and both report against the same business outcomes. Where L&D genuinely lacks workflow and data literacy, that is a build priority, and an urgent one.
“Fluency comes from knowing the work, so the business should own it.”
Also fair. Line leaders will always be closest to the work, and the best redesign ideas will come from them. The flaw is in the conclusion. Pockets of fluency emerge everywhere; without a function that owns the learning system, they stay pockets. Someone has to spot what is working in one team, codify it, and spread it across fifty. That is a learning-architecture job.
“Executives don’t feel the gap.”
This may be the most dangerous objection because the data supports it: 76% of executives say they are confident their organisation already has the right talent and skills to achieve their AI goals. That confidence sits awkwardly beside the same survey’s findings on patchy training and unclear ownership, and it means L&D cannot wait to be invited. The case has to be made in business language: capability gaps quantified, tied to problematic pilots, and priced.
“L&D earned the 7%”
This may be the most dangerous objection because the data supports it. A function that has spent years reporting course completions and satisfaction scores has trained its executives to see it as a content shop. If L&D shows up to the AI conversation with the same machinery and new labels, the 7% will persist, deservedly. The mandate will be earned by doing different work, and by measuring it differently.
Try this: Don’t score this on a spreadsheet, have a frank conversation with your leadership team and your own people about which of these four is really holding you back, then act on it this week:
The repositioning is concrete. In our work with senior leadership teams, the L&D functions that get pulled into transformation conversations have made versions of these six shifts:
Behind the table sits a harder set of questions that the business needs someone to own. What does AI fluency mean here, and how does it differ by role? Which work should be automated, augmented, redesigned, or deliberately protected? What behaviours do leaders need to model for experimentation to feel safe rather than chaotic? And underneath all of it: how do we help people navigate the fear, status anxiety, and identity disruption that AI creates? Capability building fails when people are quietly defending their sense of professional worth. The function that understands both learning and human emotion is better placed to lead this than any other in the building.
The repositioning starts inside L&D’s own team. One of our partners was holding off until top leadership agreed on an AI strategy. We advised the opposite sequence: start within the L&D function now. Building the team’s own AI fluency, running a small pilot, and learning from it firsthand gives L&D a head start, and far more influence over the strategy once leadership is ready to write it. A function that has done the work speaks with a different authority than one that has read about it.
Practice is what moves adoption, even at very large scale. Another partner, a bank with 100,000 employees, had given everyone Copilot licences and a brand-new library of AI learning content, and was still facing low adoption. We have been helping them shift the emphasis from absorbing knowledge, which ages quickly as tools and models change, to doing the work: weekly guided practice sessions in which employees apply AI to real work and learn to build agents for the workflows they own.
Crucially, our role is not only to run the sessions but to shape how the Microsoft Learning Agent itself performs. We bring learning and instructional-design expertise to bear, so that a generic copilot agent is guided, prompted, and steered to help employees build real fluency on their own work, rather than coach them generically out of the box.
The goal is for employees to build fluency on their own tasks, with guidance, on a cadence. This mirrors what the Work AI Index found across 6,000 workers: the tool itself becomes the teacher when people learn on real work, with support. Licences and content gave people access; guided practice is giving them capability.
Measurement is where the repositioning becomes visible to the CFO. Completion rates and usage dashboards describe activity. Capability metrics describe value: time-to-proficiency on an AI-augmented workflow, the share of a team able to complete a defined task to standard with AI assistance, cycle-time or quality improvement in redesigned processes. Said simply: measure what people can now do, and what it earns.
Try this: Retire one completion metric this quarter and replace it with one capability metric, defined with the business owner of the workflow it measures. One honest capability number will do more for L&D’s credibility than a year of completion reports.
L&D has spent years asking the business to take learning seriously. AI may be the moment the business has no choice. The winners of this transformation will be the organisations whose people learn fastest, whatever models they license, and AI transformation will belong to whoever owns the learning system of the organisation. If L&D defines its role as AI training, it will miss the moment. If it defines its role as organisational adaptation, it may finally become what it has always said it was: strategic.
The seat at the table is being set. It will be kept by the function that proves it can help the organisation learn faster than change happens.
Three questions to take into your next executive meeting:
If this resonates, the next step is a focused conversation about where AI capability ownership sits today, where the gaps are, and what L&D can realistically move first.
Reach out to Cori with a few lines about where your organisation is on this journey and what you are trying to change. We can then have a focused conversation about where AI capability ownership sits today, where the gaps are, and which first moves would make the biggest difference.
Cori is Principal at The DO, where she works with HR and L&D leaders to claim their seats in AI workforce transformation. She has 15 years of experience on both sides of the table, from Learning Lead at a Big Four management consultancy to working with partners across aviation, travel, FMCG, pharma, and healthcare. Connect with her on LinkedIn.
.png)
