As AI adoption expands into everyday workflows, HR leaders are discovering that the entry-level pipeline rebuild—the deliberate work of recruiting, training, and advancing early-career talent—is no longer solely a talent acquisition concern. It is becoming a knowledge strategy, one that is increasingly difficult to separate from the governance challenge AI creates. Two recent accounts illustrate why: one shows an internal audit exposing AI use that official policy had failed to catch, and the other describes the quiet collision of AI with remote work, where the transfer of organizational knowledge is now one of the hardest problems inside a business.
An Audit at Plus More Caught a Governance Gap That Had Already Opened
At Plus More, the gap between formal AI policy and operational reality became visible through an internal audit. The audit found that employees were using unregistered AI tools on work devices despite the existence of an AI policy. That pattern, commonly called shadow AI, is not always visible in official software inventories or procurement records. It forms when employees adopt tools on their own, often through personal accounts, before the organization has formally evaluated those tools for security, privacy, or reliability.
The result did not surprise Crow, the leader at Plus More who oversaw the work. The policy was new, and confidence in it among leaders was not yet high. But the audit served a specific purpose: it gave the organization a concrete evidence base to take to managers. As Crow put it, “We weren’t surprised by the result when we ran this audit. The AI policy was fairly new and leader confidence around it wasn’t high. It gave us something concrete to go to the leaders across the network and say that, even if they didn’t think their team were using AI, they definitely were and here are the risks of that.”
The significance of that statement goes beyond the detail of one company. It reveals a common failure mode in AI adoption: policies are written, distributed, and filed, but behavior continues to run ahead of governance. By the time an audit documents the gap, employees have already made independent decisions about which AI tools they trust with their work.
What is shadow AI?
Shadow AI is the use of artificial intelligence tools inside an organization without formal IT approval, registration, or governance oversight. It typically develops when employees adopt consumer-facing AI platforms or use personal accounts on work devices before the organization has evaluated the tools for security, privacy, and reliability. In the Plus More audit, shadow AI appeared as unregistered AI tools running on work machines, which meant the organization could not verify how data was being handled or whether outputs could be trusted.
The term is a direct descendant of shadow IT, which has long described unapproved hardware and software in the workplace. But shadow AI carries a sharper risk because generative AI tools can ingest large amounts of context, generate content that appears authoritative, and influence decisions with relatively little human oversight. That makes the governance question not just a compliance issue but an organizational learning issue as well.
For Kidd, the Hardest Problem Is the Transfer of Knowledge
While the audit at Plus More exposed a governance gap, another HR leader, Kidd, pointed to a deeper structural problem: AI is arriving at precisely the moment when the traditional mechanisms of workplace learning are already strained. “I think probably one of the hardest things is AI combined with the fact that we’re not all in the office. I think one of the hardest things inside an organisation now is that transfer of knowledge,” Kidd said.
That observation captures a critical change in how expertise moves through a company. In an office environment, early-career employees learned by proximity. They watched how senior colleagues framed ambiguous questions, heard how difficult conversations were handled, absorbed unwritten rules about what counted as a strong draft or a risky assumption, and gradually internalized judgment through repeated exposure. Remote work weakened that proximity. AI threatens to weaken it further by removing the struggle that once taught people how to think.
When an entry-level employee uses an AI assistant to summarize a meeting, draft a response, or generate a first version of a document, they gain speed but may lose the encounter with the underlying problem. The process of wrestling with a task—trying, failing, receiving feedback, and revising—has historically been the core of professional development. AI can collapse that process into a single step. The employee produces an answer, but the reasoning behind the answer remains invisible to them. Knowledge does not transfer in the same way.
The combination Kidd describes is more destabilizing than either remote work or AI alone. Remote work already makes incidental learning rare. AI intensifies the effect by making the absence of that learning harder to detect, because the work still gets done. A junior employee can be productive with an AI assistant while missing the deeper context that would allow them to work independently in the future. That context—why a decision was made, which information could be trusted, what was left unsaid—is precisely what an entry-level pipeline is designed to build.
Why Expanding AI Adoption Is Driving the Entry-Level Pipeline Rebuild
The phrase entry-level pipeline rebuild has begun to appear in HR strategy conversations as organizations confront the consequences of these changes. It reflects a recognition that hiring more junior staff is not enough. The entire path by which early-career people enter an organization, learn its norms, develop professional judgment, and advance toward their first promotion needs to be reconsidered for a world in which AI is embedded in daily work.
What is an entry-level pipeline rebuild?
An entry-level pipeline is the set of processes an organization uses to bring early-career people into the business and develop them into productive professionals, from sourcing and hiring through onboarding, training, mentorship, and early promotion. A rebuild of that pipeline means redesigning those processes for new conditions—including hybrid work, shifting skills expectations, and the presence of AI in daily workflows—rather than simply hiring more junior staff.
The connection to Kidd’s concern is direct. If knowledge transfer is the hardest problem inside an organization, then the entry-level pipeline is one of the most important places to solve it. That pipeline is, at its core, a structured version of knowledge transfer. It exists to ensure that newcomers acquire not only task competence but also context, judgment, and trust. When AI changes how work is performed, that pipeline must change too.
What makes this particularly difficult is that AI does not affect all parts of the pipeline equally. Recruitment can be redesigned with assessment tools and realistic job previews. Onboarding can be updated with structured content. But the informal, unstructured parts—the side conversations, the quiet observation, the experience of making a mistake in a low-stakes setting and learning from it—are much harder to rebuild. Those parts depend on human interaction, and AI cannot substitute for them. It can only make them rarer unless deliberately protected.
Leader Confidence Was the Missing Ingredient in the AI Policy
Part of the Plus More story is about the relationship between policy and leadership confidence. The policy was new, and the leaders responsible for enforcing it were not yet sure what it covered or how to talk about it. That combination created an environment in which employees made their own choices about AI tools, and managers were not equipped to notice or respond.
Crow’s description of the audit is telling: leaders across the network did not think their teams were using AI, but the audit showed otherwise. This is not simply a failure of detection. It is a symptom of how AI adoption has evolved. Employees have been experimenting with generative AI since the technology became widely available, often without formal training or approval. By the time an organization publishes a policy, many employees have already integrated those tools into their workflows. The policy is meant to direct behavior that has already formed.
The experience at Plus More suggests that the presence of a policy is not the same as shared understanding. Leaders who lack confidence in a policy are unlikely to raise the subject with their teams, model the recommended behavior, or challenge questionable usage when they see it. The policy becomes a document rather than a practice. That is why Crow’s account of the audit emphasizes the concrete moment it created: for the first time, leaders had a reason to talk about AI with their teams, and they had specific risks to point to.
What emerges from this sequence is a governance model that is less about policing tools and more about building shared awareness. The audit did not simply produce a list of violations. It produced an opening for conversation. The learning that followed was not a one-time announcement but an ongoing effort to communicate the same messages repeatedly and in different formats.
Critical Thinking Training Now Sits Alongside Copilot Training
The response at Plus More was not a blanket restriction on AI tools. Instead, the firm added a second stream of training focused on critical thinking and on the conditions under which AI output cannot be trusted. That stream runs alongside practical tool training on Copilot. The two-track structure reflects a broader shift taking shape across HR: AI readiness is no longer only about learning how to use a tool. It is about learning how to evaluate the tool’s output, when to question it, and when to keep a human in the loop.
The distinction between tool training and judgment training is important. Tool training teaches an employee how to use a system: how to prompt, how to interpret a response, how to integrate the output into their work. Judgment training teaches an employee when not to use the system, or at least when to be suspicious of what it returns. Both are necessary, but they address different abilities. A person can be highly skilled at using an AI tool while still lacking the critical judgment to know when the tool is wrong.
That point is particularly relevant for entry-level employees, who have less experience to draw on when evaluating AI-generated content. A senior professional can often recognize a plausible but subtly flawed answer because they have seen similar problems before. A junior employee may not have that baseline. If organizations want AI to enhance entry-level work rather than erode it, they need to build the critical judgment layer before the tool training becomes routine.
“We’ve learned that having a policy is not enough; teams need to hear the key messages clearly, repeatedly, and in different ways,” Crow said. That sentence could serve as a compact design principle for the next phase of AI adoption. Policies create direction, but they do not create competence. Competence comes from repetition, from reinforcement, and from opportunities to practice judgment in real situations.
The second training stream at Plus More is notable because it treats critical thinking as a trainable capability rather than an innate trait. It assumes that employees can improve at identifying unreliable AI output through structured practice. That assumption is supported by the way professional judgment develops in other domains, where novices learn to spot patterns of error through exposure and feedback. In an AI-augmented workplace, that kind of feedback becomes an essential part of the entry-level experience.
The Next Phase of AI Adoption in the Workplace
The experiences described at Plus More and by Kidd are early signals of a broader transition. Organizations are moving from a phase of experimentation with AI to a phase of institutionalization. In that phase, the central question is no longer which AI model to deploy or how much productivity it can unlock. The central question is how AI changes the way knowledge is created, stored, and passed on.
That question has direct consequences for the entry-level pipeline rebuild. If early-career employees learn by doing, then AI alters the curriculum. If they learn by observing others, then remote work alters the classroom. If they learn through feedback, then organizations need to be deliberate about creating feedback loops that AI cannot replace. The challenge is not to reject AI but to design work so that the parts that build human judgment remain intact.
The Plus More audit shows that employees will continue to use AI tools regardless of formal policy. The question is whether organizations can create enough trust, transparency, and critical engagement to make that use constructive rather than haphazard. The training model that emerged from the audit suggests one direction: combine practical tool proficiency with repeated, reinforced messages about the limits of AI output, and place judgment at the center of the curriculum.
Kidd’s concern about knowledge transfer points to a deeper risk. In a distributed, AI-assisted workplace, the informal pathways by which expertise moves from one generation of employees to the next may break. The tools may become more capable, but the people using them may become less able to reason independently about what they produce. That is not an argument against AI adoption. It is an argument for treating knowledge transfer, entry-level development, and AI governance as parts of the same problem.
As HR leaders continue the entry-level pipeline rebuild, the most durable strategies will likely be the ones that recognize AI as both a productivity tool and a test of organizational learning. The technology will keep expanding. The organizations that manage it well will be those that remember what entry-level employees have always needed: exposure to real complexity, feedback from experienced people, and the space to make mistakes while the stakes are still low. AI can support that process, but it cannot provide what only human interaction has ever delivered.
