Organisations are moving quickly to embed AI and agentic systems into their workflows, but one crucial factor is slowing adoption and improving outcomes: confidence in the experience itself. In Pendo’s recent UK study of 1,000 adults, 72% said AI is exciting and full of opportunity, but over two-thirds (69%) reported that, whilst it is useful, it’s often frustrating to use.
This mismatch between AI’s potential and experience is creating an AI experience gap: the difference between how people expect AI tools to act and how they actually behave in real-world contexts. This gap has become the biggest bottleneck between investment and impact. Without addressing it, organisations risk deploying systems that look powerful on paper but fail in practice.
The AI experience gap matters more in agentic workflows
As organisations shift toward more autonomous, agentic systems, this experience gap will become even more significant.
In traditional software, friction slows people down. Yet in agentic systems, friction can halt adoption entirely. When an AI agent performs a task incorrectly, misunderstands intent, or provides an unexpected response, users quickly lose trust.
Research shows how quickly this trust erodes in practice: Almost all (95%) of regular AI users spend time each week re-prompting tools just to get a better answer. 38% switch back to search engines when AI struggles, and 16% give up altogether. These behaviours are small on their own but significant at enterprise scale, especially when agents power high-volume, high-risk, or customer-facing tasks.
If organisations don’t actively study how employees interact with agents, where they break down, or where they need more support, agents won’t scale, regardless of their technical prowess.
The gap between how AI should work and how it actually works is widening, and without intervention, it will cap the value organisations can extract from AI.
Confidence also plays a central role in AI’s success. Only 37% of users firmly trust AI outputs, and fewer than half (46%) feel confident in their ability to get the best results from AI tools. Low trust drives low adoption, and low adoption limits ROI, especially in industries where compliance, auditability, and predictability are critical.
The hidden friction that’s slowing adoption
Agentic AI has overturned how work gets done. Executives often guess wrong about how people will actually use AI agents. Companies track basic numbers like how many people use them, how often they break, or how fast they run. But most can’t answer the bigger question: Are these agents actually helping people get more done?
Real adoption problems surface when users get confused, backtrack, re-prompt, or skip AI entirely. These are the moments that determine if AI becomes a trusted part of work, or a tool people quietly abandon.
Measuring AI experiences: the missing foundation
To reduce the AI experience gap, organisations need genuine visibility into how people interact with AI tools. It’s not enough to know that a model performs well in testing, or that an agent has been added to a workflow. What matters is what happens when people hesitate, when they switch tools, why they lose trust, and how they attempt to recover when AI falls short.
Because people interact with agents through conversation, organisations can uncover trends via what users are actually saying: the problems they’re trying to solve, the moments of confusion, and the behaviours that signal friction. In many cases, users are already telling companies how they want the agent to behave. It’s up to organisations to listen.
With this level of insight, organisations can fix confusing workflows, improve prompts and in-app guidance, retrain users who are unintentionally misusing tools, and introduce guardrails that prevent errors. Just as importantly, teams can establish a continuous feedback loop, asking users how the agent’s experience is evolving as improvements are made.
This allows enterprises to measure the actual impact of agentic automation on productivity, quality, and time saved.
Why training and guidance now matter more than model size
This research reveals something organisations can no longer ignore: people want help using AI better. 86% of respondents said companies should provide training or guidance on how to use AI tools effectively.
Many organisations are using more and more AI, but are leaving employees on their own, stuck trying to figure out how these new tools work. This creates the impression that AI is intuitive by default, when in reality, good AI usage is a learned skill. People need to understand when to trust AI, when to challenge it, and how to guide it.
Without this foundation of knowledge, even the most powerful agentic systems will underperform. Training cannot be a one off webinar or a static set of guidelines. Instead, teams need a mix of hands-on support, ongoing education, set guardrails, and contextual guidance that adapts to their real behaviour. This will accelerate adoption, and ultimately shift AI to being used by default.
The responsibility of AI is shared
Interestingly, users believe that responsibility for reliable AI does not rest solely with the team that build the AI agent. When asked who should ensure AI tools work, responses were split:
- 33% say AI vendors
- 25% say all parties equally
- 10% IT teams
- 10% regulators
- 9% end users themselves
This distribution is important. People understand that AI is a shared system of responsibility, where model developers, deployers, and organisations all play a role in its success. Enterprises buying these tools can’t assume that responsibility ends at procurement or initial deployment. They must actively shape how AI is used, monitored, and improved across the business.
Without this involvement, AI adoption becomes fragmented, and in regulated or risk-sensitive industries, this shared responsibility is even more critical. AI’s results must be clear, trackable, and provable. When employees trust the system, the company gets stronger. If people don’t trust AI, rely on workarounds, or revert to manual processes, the organisation loses the very efficiencies AI promised to unlock.
To close the AI experience gap, get closer to your users
The companies that win with AI will be the ones that focus on how people actually use it, not just the technology itself. When employees trust their AI tools and get the support they need, adoption goes up and productivity improves. Companies that monitor how AI is working for their people (not just whether it’s deployed) can spot problems early and fix them fast. This creates a real advantage.
The biggest gap between AI investment and actual results is the end-user experience. Close that gap by continuously learning what works, designing for real workflows, and helping people succeed. Companies that treat AI as an experience to manage will see higher adoption, stronger performance, and clearer ROI. Those that treat it as just another tech deployment will stay stuck in pilot mode.
