Skills, trust, and uptake: why people still don't use generative AI

Is this a business, management, or employee problem?

The contrast between the universal awareness of generative AI and actual daily usage is one of the most striking trends in the current tech landscape. This usage difference is a structural divide that risks leaving significant portions of the workforce behind. 84% of younger digital natives use AI at least monthly compared to 37% of baby boomers. This suggests that innate tech literacy (often attributed to younger workers) isn't enough to drive adoption across a multi-generational workforce. If an organisation's AI strategy relies on employees figuring it out in their own time, it's effectively excluding a huge portion of its talent pool. Management and business culture should encourage AI integration into their teams and processes with adequate training on how to get the most out of the technology.

Pilot purgatory occurs when a firm uses AI in an initial testing phase but fails to scale it to full production, despite major initial investments. This is usually for four reasons:

Companies get stuck because they treat AI as a standalone feature rather than a fundamental shift in how work is performed. The firms that break through this ceiling are those that move beyond the basic chatbot phase. They stop asking "What can this tool do?" and start asking "Which of our existing bottlenecks can this tool solve?" looking to tasks like workflow automations, data analysis, or compliance checks. This shift in perspective is the difference between a novelty and a long-lasting competitive advantage.

So what's preventing teams from asking more of the AI that's available to them?

A critical reason that users aren't moving out of this pilot mindset is a lack of trust. Hallucinations and inaccuracy remain a limitation of generative AI, and 72 per cent of AI-sceptics are worried about AI hallucinations. For companies where brand trust is critical, the risk of providing inaccurate information via AI may be unpalatable. For such companies, fact-checking and cross-referencing AI's efforts are essential, but this can be incredibly time-consuming. If the perception amongst a team is that the output of a tool takes longer to fact-check than to write from scratch, its perceived usefulness drops. These teams may lack the time, knowledge, or incentive to discover how to improve the accuracy of AI outputs, leaving them stuck with unsatisfactory defaults.

Governance is another major component of the adoption of AI tools. Where workplaces are slow to issue sanctioned tools or guidance for AI usage, shadow AI crises can develop, where people use AI in secret. 72 per cent of UK employees have used unapproved AI tools, risking data leaks and other security concerns. When there's no sanctioned AI-use permitted, other workers will simply avoid AI entirely to avoid falling foul of HR. This can cause a lag in productivity, slower work, or manual errors, making businesses unnecessarily struggle when compared to their competitors. This lack of visibility creates a culture of apprehension that stifles innovation.

Finally, the fear of AI displacing jobs hasn't gone away. According to our Digital Etiquette: Unlocking the AI gates research, 44 per cent of 18–25-year-old workers feel that AI is likely to replace jobs. This fear breeds cynicism. If workers feel pressured to keep up with AI-driven productivity demands without being given the tools, training or time to adapt, their job satisfaction plummets, and the likelihood of them adopting AI lowers, too.

Just having access to tools doesn't mean you can use them effectively. We've moved past the era where simple prompting was enough for you to get ahead. Now we have agentic AI (where AI acts as a digital teammate capable of automating complex, multi-step workflows), so the technical foundation that someone needs to truly harness AI's best capabilities is much more sophisticated. If you feel that your team has fallen behind, however, it isn't too late to turn things around.

A lack of formal training remains the biggest bottleneck. Our Digital Etiquette: Unlocking the AI gates report found that 46 per cent of knowledge workers who had over 20 hours of AI training saved at least 11 hours of work tasks a week. Contrast that with those who've had no training; they're often stuck in a cycle of distrust where one bad experience (like a hallucinated fact) stops all future experimentation. Effective training should include AI literacy: understanding where the AI gets its information, its limitations, and how to verify its output against trusted sources. Many workers are self-teaching during their personal time, leading to a stark disparity in confidence levels, but a higher understanding of upcoming possibilities.

For example, in the Atlassian ecosystem, we're seeing more self-directed AI experimentation with tools like Atlassian's Rovo. Rovo is an AI-powered teammate that can search across disparate software stacks to find information and help teams automate tasks. To use a tool like Rovo as effectively as possible, a worker needs to understand the possibilities and the limitations, such as how their organisation's data is structured, but importantly, they need an environment that encourages continued experimentation.