New Harris Poll research for Box points to the need for clearer guidance, better workflow integration, and more thoughtful AI rollouts
Every day, knowledge workers are challenged to apply an ever-growing number of AI tools to an even larger number of business tasks. How do they feel about it?
Last month the Harris Poll on behalf of Box surveyed approximately 800 U.S. employees who already use AI for work. The surprising top-line finding: a full 61% of respondents are maxed out on AI, wanting the technology’s help with the same number of tasks or fewer than they use it for today. Only 39% actually want AI to help them with more tasks.
For IT leaders, this finding highlights the importance of learning what is and isn’t working with today’s AI integrations in order to make tomorrow’s agentic workplaces as effective as possible.
“People may have reached the limits of what they know how to do with AI today,” says Box CIO Ravi Malick. “They may need help understanding what else is possible. Or they may need access to information their tools can’t use yet. We need to understand what’s holding them back before we as IT leaders decide how to expand.”
Key takeaways:
- 61% of workers want AI’s help with the same number of tasks or fewer; 39% want AI to help with more tasks.
- 57% say having too many options makes them less likely to use AI.
- 78% would use AI more if they could access more capabilities without having to understand differences between tools or models.
Options without context can discourage AI use
For knowledge workers, access to AI comes with lots of decisions: which tool to open, what data to give it, and whether the answer they get back for any given query is useful enough to include in their work. These choices can feel overwhelming; 57% of AI users say having too many options makes them less likely to use the technology.
For teams that are investing in broader access, these numbers are a clarion call to examine how they introduce and organize those choices to prevent decision paralysis, and to consider whether employees can readily identify an approved tool for the work in front of them without getting overwhelmed.
38% of survey respondents say they often or always have to repeat instructions, re-enter information, upload the same content, or recreate context when using multiple tools.
AI users also face the burden of continually supplying context. 38% of survey respondents say they often or always have to repeat instructions, re-enter information, upload the same content, or recreate context when using multiple tools.
An AI tool might produce an impressive answer once it has all the information it needs. The pertinent question is how much work a human must do to gather that information. The findings suggest people might be more likely to use AI if it provides or is connected to the context and data sources it needs in order to do seamless, accurate work.
More capability should be easier to put to work
The opportunity before enterprises is substantial but comes with conditions. 78% of workplace AI users say they’d use the technology more if they could access additional capabilities without having to understand the differences between tools or models.
They also want improvements to fit into their working routines. 85% agree their organization should be able to change AI models without disrupting how employees work. As capabilities evolve, employees should benefit without continually having to relearn how to use a tool.
40% of workers reported wanting to choose the tool or model for each task themselves.
And it’s worth clarifying that simplifying the AI choice experience doesn’t mean making every decision for employees. When asked how they’d prefer to navigate different AI tools, 40% of workers reported wanting to choose the tool or model for each task themselves. This answer — the most popular response to that question — suggests the wisdom of offering clear defaults and recommendations for everyone, while also providing capabilities for those intrepid employees who do want to make their own selections.
Three questions before expanding AI use
At Box, we required baseline training and education before employees received access to enterprise AI accounts. Colleagues opted in more rapidly as they saw others using the tools productively, coupled with repeated communication, transparency, and pathways for people to develop new skills spearheaded by leadership.
But Malick sees education as only part of the answer; IT leaders also need to reduce the number of technical decisions employees have to make in order to get useful results. His team is exploring how employees could work through a familiar interface while requests are routed to different AI models behind the scenes, based on the task.
Employees want to use AI to do their jobs, but they don’t want to evaluate every model before they can get started.
“Employees want to use AI to do their jobs, but they don’t want to evaluate every model before they can get started,” he says. “Part of our responsibility as IT leaders is to put the architecture in place so they can use an effective model for the task without having to make that decision every time.”
What might that architecture look like? Here are a few key questions to ask before extending AI into more tasks:
Which task do employees want improved? Ask users where today’s assistance falls short and where additional help would matter. Give them room to identify work they want to retain themselves.
What extra work does using AI create? Follow a task through finding the right tool, supplying company information, reviewing the answer, and using the result. Look for repeated effort and awkward handoffs that make a potentially useful capability harder to apply.
Which choices and skills do people need? Find out who wants direct control over tool selection, who would welcome a recommendation, and what training each group needs. Test the experience with employees before expanding it.
IT leaders should also study where their companies’ current AI experience falls short. Helping employees find the right tool, connecting them with information they need, and making sure everyone can study the results will enhance existing investments and reveal where additional capabilities would help. To learn more about how those goals are realized in practice, read Box COO Olivia Nottebohm’s account of experimentation, education, and implementation in How AI transformation really happens.
Survey methodology
This research was conducted online by The Harris Poll on behalf of Box from October 7–8, 2026. The survey included 2,016 U.S. adults, including 961 knowledge workers who primarily work in a professional setting and use a computer as part of their job. Of those knowledge workers, 818 use AI tools or features for work-related tasks. Data are weighted where necessary by age, gender, region, race/ethnicity, household income, education, marital status, size of household, and political party affiliation to bring them in line with their actual proportions in the population.
