When Everyone Uses AI, Who Still Understands the Work?

The competency erosion problem: Small Businesses & Nonprofits

Recent research suggests that concerns about AI weakening human skills are not abstract. In a 2026 SHRM report on AI in the workplace, workers widely reported both increased use of AI tools and rising unease about how that use affects the quality and originality of their work. Nearly three in five workers surveyed by SHRM said they had worked with a colleague whose actual capabilities did not match the skills they were supposedly hired for, highlighting a mismatch between paper qualifications and observed competence. A similar share — around 57 percent — said they were at least moderately concerned that heavy reliance on AI was leading to lower‑quality work that had not been thoughtfully reviewed or vetted.[1][3]

Those worries align with broader research. Strategy analysts at BCG have warned that when organizations lean too hard on AI for knowledge work, they risk eroding critical cognitive skills such as problem solving, judgment, and deep analysis. Academic work has raised similar alarms: a 2024 theoretical paper on AI assistance argued that automated support can accelerate skill decay and hinder skill development, particularly when users stop engaging with the underlying tasks and instead become passive overseers of machine output. A 2023 study on AI and workers’ skills emphasized that without deliberate upskilling and reskilling efforts, AI integration is likely to widen gaps in transversal skills such as critical thinking and adaptability.[2][4][5]

What makes this “competency decline” distinct from classic disengagement or burnout is its mechanism. Instead of people doing less work, people may still produce many deliverables — they just do so with less understanding, weaker reasoning, and fewer original contributions. Over time, that hollows out the very capabilities organizations rely on to detect errors, manage risk, and adapt to change.[6][7][8][2]

Why small businesses should care first

Large enterprises have more redundancy: if one team’s skills atrophy, there are often other teams, formal training programs, and deep specialist benches to catch problems. Small businesses, by contrast, typically rely on a handful of key people handling multiple functions — operations, customer service, basic finance, and HR — often with no dedicated L&D infrastructure. In that environment, AI‑driven competency decline is not a theoretical risk; it can immediately affect customer experience, compliance, and financial performance.[9][10][11][12][2]

Small‑business technology surveys show that AI and automation tools are now common even in firms with fewer than 50 employees, where owners frequently deploy them to “do more with less” amid rising costs. SHRM’s State of the Workplace and AI research likewise finds that leaders expect AI integration and AI‑specific upskilling to accelerate, but many organizations have not yet built mature governance structures around these tools. That leaves small employers in a particularly fragile position: they gain efficiency from AI, but lack clear policies, training programs, or metrics to ensure those tools do not quietly weaken human skill over time.[13][2][14][15][3][9]

From a risk standpoint, that matters for several reasons:

· Error detection and compliance. Small teams often manage payroll, scheduling, and basic legal compliance themselves. If employees stop understanding the logic behind AI‑generated schedules, classifications, or notices, they are less likely to catch mistakes that could trigger wage‑and‑hour violations or other legal exposure.[10][11]

· Customer trust. Many small businesses compete on quality and relationships, not scale. If AI‑generated emails, proposals, or recommendations are not truly understood or checked by staff, clients will eventually notice generic or inaccurate work — and may leave.[6][9]

· Succession and resilience. When a key employee leaves or is unavailable, a small business needs others to understand critical decisions, not just how to press “generate” again. If AI has masked skill erosion for years, transitions become much riskier.[2][8]

In short, small businesses cannot afford to treat AI competency decline as a vague future issue. It is a near‑term operating risk that deserves the same attention as cybersecurity, financial controls, or safety protocols.[12][10]

Evidence that skill erosion is already happening

Several data points suggest that skill erosion linked to AI is not just hypothetical.

SHRM’s 2026 AI research reports indicate that most HR professionals believe repeated drafting, editing, or refining work with AI is at least moderately likely to reduce original thinking. In parallel, workers themselves are noticing the shift: roughly seven in ten workers surveyed said critical thinking is more important now because of AI, implying a sense that the tools make it easier to produce content but harder to distinguish strong answers from weak ones.[1][14]

Other surveys have found that a notable portion of workers — 39 percent overall and nearly half of Generation Z in one recent poll — believe heavy reliance on AI has already weakened their own skills. Research teams in Europe have warned that overdependence on AI to guide decisions can hollow out expertise, especially when organizations fail to preserve practice tasks that require human judgment. All of this aligns with broader human‑factors literature suggesting that automation can produce “skill fade” when people no longer engage deeply with procedures they once had to perform manually.[7][4][5][8]

Taken together, these findings build a consistent picture: AI can boost productivity, but without safeguards, it also nudges workers into roles where they supervise outputs they do not fully understand. That is competency decline in practice.[6][2]

Why policy is the right response (not just training tips)

Many employers try to address AI risks informally — telling staff to “double‑check AI outputs” or encouraging people to “still think for themselves.” Those messages help but are insufficient. Competency decline is driven by system design and incentives, not just individual intentions. Without clear policies, workers may feel business pressure to prioritize speed and volume over understanding, especially in small businesses where everyone is stretched thin.[9][16][6][5]

Formal AI policies are needed for at least four reasons:

1. Clarifying expectations. When a written policy defines what “responsible AI use” means — where AI can be applied, where it must be avoided, and how human review works — it reduces ambiguity and gives employees concrete guardrails.[1][17]

2. Aligning incentives. If small businesses only measure productivity (e.g., number of tickets closed or documents produced), workers will naturally use AI to maximize volume, even when quality or capability suffers. Policies can embed expectations about work quality, error rates, and demonstrated skills, shifting incentives away from speed alone.[2][1]

3. Protecting developmental work. Research on skill erosion stresses that organizations must protect tasks that build expertise — diagnosis, reasoning, design — not just those that produce output. A policy can explicitly identify which tasks must remain human‑led and practiced regularly, even if AI could technically perform them.[18][5][8][2]

4. Supporting compliance and trust. As regulators and customers become more aware of AI risks, organizations will be expected to show how they govern these tools. Clear policy language around data sensitivity, bias risks, and review procedures builds external trust and reduces legal exposure.[14][3]

For small businesses, these policies need not be long or academic. They do, however, need to be explicit, applied consistently, and linked to performance management so that they influence daily behavior rather than remaining aspirational statements.[11][19]

Core elements of an AI‑competency policy for small businesses

Based on current research and emerging HR guidance, small employers can structure an AI‑competency policy around five practical pillars.[1][2][5]

1. Define AI literacy as judgment, not just tool use

SHRM’s AI reports emphasize that workers need to understand principles of responsible use, data sensitivity, and verification — not just how to prompt a model. Academic studies on skills similarly stress transversal capabilities such as critical thinking and adaptability as central to long‑term performance.[1][14][5]

For small businesses, this means:

· Requiring employees to be able to explain how AI produced a given answer (at least in conceptual terms) and why that answer makes sense in context.

· Including AI literacy and judgment in performance reviews: workers should be assessed on whether they use AI to enhance understanding, not to avoid it.[2][15]

· Providing short, scenario‑based training showing when AI is appropriate, when it is risky, and how to cross‑check outputs with source data or domain logic.[6][1]

2. Protect core practice tasks in every role

Research on automation and skills recommends mapping which tasks build expertise and deliberately preserving those tasks in workflows. Small businesses can:[18][8]

· Identify “developmental tasks” in each role — examples include troubleshooting unusual customer issues, designing a process, or interpreting financial trends.

· Explicitly forbid “one‑click automation” of these tasks: AI can assist with options or drafts, but employees must perform and document their own reasoning steps.[2][5]

· Ensure junior staff get hands‑on practice, not just final review of AI‑generated work, to prevent a generation of employees who have never built skills directly.[7][18]

3. Make AI assistants visible in the org chart

When AI is invisible, accountability blurs. Workers may assume “the system” is responsible for errors, not them. SHRM and other advisory bodies recommend naming AI roles and responsibilities, even if they are tools rather than people.[6][1][15][8]

In a small‑business context:

· List key AI tools (e.g., drafting assistant, scheduling optimizer, analytics copilot) alongside the human roles they support.

· Clarify that humans own outcomes: the person whose job is supported by AI is still accountable for the decision or deliverable.[2][14]

· Use this mapping to guide training and audits: if a tool affects customer communication or compliance, those areas warrant extra oversight.[10][12]

4. Redefine metrics: beyond “more, faster”

BCG and other sources note that AI makes it easy to improve raw productivity metrics while obscuring declines in capability. SHRM’s research similarly urges leaders to track how AI affects quality, not just output.[6][1][2][15]

Small businesses can adjust metrics by:

· Tracking error rates, rework, and customer complaints, especially in areas where AI is heavily used. If errors rise while volume rises, competency may be eroding.[20][9]

· Measuring cycle time alongside quality: AI should help people spend more time on deep work, not only compress timelines at the expense of thoughtfulness.[2][5]

· Recording capability indicators such as skills demonstrated in real projects, not just completion counts. For instance, noting when employees successfully handle non‑standard cases without heavy AI dependence.[18][3]

5. Build a simple review and escalation path

Studies on AI and risk argue that organizations need clear processes for reviewing AI‑influenced decisions, especially when stakes are high. For small businesses, this can be minimal but meaningful:[6][8]

· Require a second‑person review for AI‑generated outputs in high‑risk areas (e.g., legal notices, sensitive customer communications).

· Create a simple escalation rule: if an employee does not understand an AI suggestion, they must seek clarification or override it rather than accepting it by default.[1][2]

· Periodically sample AI‑assisted work to check whether employees can explain their reasoning and identify alternatives — not just repeat what the tool produced.[4][5]

How HR and owners can implement this without big budgets

Small businesses rarely have the luxury of large policy teams. Still, research and practice show that modest, well‑targeted steps can materially reduce AI‑driven competency decline.[13][5]

Practical starting points include:

· Update the employee handbook. Incorporate a short AI section defining responsible use, prohibited uses, and expectations for review and judgment. HRdeck guidance highlights that handbooks, when updated and clear, are powerful tools for aligning behavior and reducing liability.[19][21]

· Add AI literacy to onboarding. New hires should learn how the business uses AI, which tasks are preserved for practice, and how their performance will be evaluated in relation to these tools.[1][15]

· Run micro‑drills. Once a quarter, pick a common AI‑assisted task — such as replying to a customer, drafting a policy note, or analyzing a small dataset — and ask staff to perform it with minimal AI assistance, then compare results.[2][18]

· Review roles most exposed to AI. Use simple mapping (who uses which tools, for which tasks) to identify where the risk of skill erosion is highest, then prioritize those areas for policy, training, and oversight.[5][1]

These steps do not eliminate AI; they reposition it as an amplifier of human skill rather than a replacement. That re‑positioning is vital for maintaining the competence small businesses rely on to survive in volatile markets.[9][3]

Final thought: AI as a tool, not a crutch

Across research from SHRM, consulting firms, and academic studies, a consistent theme emerges: AI is powerful and here to stay, but the organizations that will thrive are those that treat human capability as a non‑negotiable asset. For small businesses, that means adopting AI enthusiastically — but with explicit policies that preserve original thinking, critical judgment, and practice in core skills.[1][2][5]

If workers can quickly show you an answer but cannot explain why it is correct, capability is at risk. Clear AI policies, thoughtful metrics, and deliberate skill development are how small employers ensure that “smart tools” do not slowly make their workforce less smart over time.[2][4]