Generative AI shifts the question from “who has access” to “who knows how to use it well”?

Generative AI is entering enterprise environments at a pace that is forcing organisations to rethink not only tooling, but workforce strategy, governance, and competitive positioning. Research from Zhe Zhu (University of Vaasa) reframes a commonly stated concern: The risk is not simply that artificial intelligence will displace workers, but that organisations and employees who fail to adopt it effectively may fall behind peers who do.

At the centre of the research is a practical observation. This is where employees who treat generative AI systems such as ChatGPT or Gemini as collaborators rather than adversaries show higher levels of engagement, adaptability, and confidence in their roles. For enterprise deployment, this finding shifts the focus from technology selection to behavioural adoption and organisational design.

A Workforce Divided by Adoption Capability

The spread of generative AI has triggered familiar anxieties around automation and job displacement. In many organisations, these concerns appear as resistance to adoption, reluctance to experiment, or over-reliance on manual processes even where augmentation is possible.

Zhu’s research suggests a more nuanced outcome: concern about AI can act as a catalyst for capability development. Employees who perceive risk often respond by acquiring new skills, learning how to prompt effectively, validate outputs, and integrate AI into their workflows.

Cementing the academic output with tech business reality, NVIDIA CEO Jensen Huang has framed this dynamic in stark terms. This is that workers are not being replaced by AI directly, but by colleagues who know how to use it more effectively. In enterprise settings, this translates into a widening productivity gap within teams, where early adopters begin to outperform peers in knowledge work, analysis, and decision speed.

Canadian enterprises offer early evidence of this divide. Shopify, for example, has encouraged widespread internal use of generative AI tools to draft content, support coding, and accelerate decision cycles. The company’s internal messaging has emphasised AI as a baseline expectation for productivity rather than an optional tool. By contrast, organisations that restrict access or fail to provide structured training often experience fragmented usage and inconsistent outputs.

Trust as an Operational Variable

A critical variable identified in the Vaasa research is trust. Specifically this is how much employees rely on AI outputs and how critically they engage with them.

Two failure modes are emerging in enterprise deployments. These are:

• Over-trust, where employees accept AI-generated outputs without verification, leading to quality, compliance, or reputational risks
• Under-trust, where employees dismiss AI outputs outright, forfeiting productivity gains and insight generation

Managing this balance is becoming a core governance challenge. Financial institutions in Canada, such as RBC and TD, have approached this by embedding AI usage within structured controls. These include model validation frameworks, the human-in-the-loop review processes, and use-case restrictions based on data sensitivity.

In practice, trust becomes an engineered outcome rather than an individual preference. Organisations must define where AI can be relied upon, where it must be verified, and where it should not be used at all.

Enterprise Deployment: From Experimentation to Operational Integration

Many organisations have already passed the initial phase of experimentation—pilot projects, innovation labs, and isolated use cases. The next phase is more demanding: integrating generative AI into core business processes.

Zhu’s work points to the importance of structured implementation, supported by governance, ethics, and alignment with business objectives. This reflects a broader shift visible across Canadian enterprises, where AI adoption is moving from IT-led initiatives to board-level strategic priorities.

Examples include banks, telecom companies, and public sector organisations, which are now formalising AI deployment through centralised AI operating models, together with cross-functional governance committees and the defined ownership of AI-enabled processes.

This transition requires a clear framework. Zhu proposes an eight-step model that guides organisations from initial experimentation to embedded use. While the specific steps are not detailed in the summary, the principle aligns with what is already observable in enterprise practice. This is, AI must move from ad hoc usage to repeatable, auditable, and measurable processes.

One of the more subtle implications of the research is the distinction between AI as a tool and AI as part of a workflow. Many organisations initially deploy generative AI as a standalone application—accessible through chat interfaces or isolated integrations.

More advanced adopters are embedding AI directly into workflows. This includes drafting reports within document management systems. A further example is generating code inside development environments. AI is also supporting decision-making within enterprise resource planning (ERP) tools.

Canadian firms in sectors such as mining, energy, and insurance have begun integrating AI into operational workflows, particularly for documentation, compliance reporting, and predictive analysis. In these cases, AI becomes less visible as a separate system and more embedded in daily work.

This shift alters how productivity gains are measured. It is no longer a question of whether employees are “using AI,” but how their output changes as AI becomes part of standard operating procedures.

Managing Ethical, Legal, and Data Risks

Enterprise deployment introduces risks that are absent or less visible in consumer use. These include:

• Data leakage through prompts
• Use of proprietary or sensitive information
• Bias and inaccuracies in generated outputs
• Regulatory exposure

Organisations must also navigate emerging regulatory frameworks, including federal discussions around the Artificial Intelligence and Data Act (AIDA). Even in its evolving state, this signals a future in which companies must demonstrate structured governance over AI usage. Zhu’s research reinforces the need for organisations to address these concerns as part of implementation, rather than as an afterthought. Practical measures include:

• Clear policies on acceptable AI use
• Controlled environments for sensitive tasks
• Audit trails for AI-assisted decisions
• Employee training in critical evaluation of outputs

In sectors such as healthcare and life sciences, where data integrity and validation are already tightly controlled, these requirements align naturally with existing GMP and quality frameworks. In other industries, they represent a new layer of operational discipline.

The Path Toward AI-Native Organisations

The research suggests that organisations are moving toward a state where AI is embedded across processes and decision-making. This does not imply full automation, but rather a hybrid operating model in which human judgment and machine-generated insight are interdependent. Early indicators of this shift include AI-assisted decision support in executive planning. To this can be added automated generation of first-draft outputs across functions and real-time analysis embedded in operational dashboards.

In practice, Telus have already invested heavily in AI across customer service and network optimisation, demonstrating how AI can move beyond isolated use cases into system-wide capability.

The transition to this model requires changes in workplace skills (including prompting, validation, and interpretation). It is also reshaping roles, placing greater emphasis on oversight and synthesis. This trajectory is also altering metrics in terms of measuring output quality and speed rather than effort.

Economic Impact and Emerging Opportunities

Zhu situates generative AI within a broader industrial shift, where new categories of work emerge alongside automation of existing tasks. This pattern is already visible in Canada, particularly in areas such as data centre expansion in provinces like Quebec and Ontario. For such enterprises, this creates both an opportunity and a challenge. While some roles may diminish in importance, others will gain prominence. The key variable is how quickly organisations invest in reskilling and capability development. There is also a strategic dimension where companies that build internal expertise in AI deployment, governance, and integration are better positioned to adapt as the technology evolves.

The central message emerging from the Vaasa research is that generative AI does not produce uniform outcomes across organisations. Its impact depends on how it is adopted, governed, and integrated into work. This presents a call for enterprises to focus on developing employee capability and embedding AI into workflows.