The technology industry is sprinting from copilots toward autonomous agents that can execute tasks across apps, make decisions, and increasingly act with limited supervision. eWeek recently described that evolution as an enterprise shift from AI copilots to autonomous workflows, as companies look to move AI from suggesting what to do to actually doing it.
But Daily Tech Insider readers — working professionals and tech-minded consumers alike — seem considerably less eager to hand over the office keys.
Across seven DTI Pulse Checks from March through August 2026, respondents repeatedly welcomed AI assistance while resisting AI authority. The polls covered workplace agents, personal computers, everyday chores, medicine, weather forecasting, and even the prospect of reporting to a machine.
The clearest example came after DTI covered an experimental AI store manager that helped fire a human employee — though only after researchers essentially had to spoon-feed the forgetful bot its own employee handbook. Asked on Aug. 17 whether they would work for an AI boss, 85.4% of respondents said no. Another 13.4% said maybe, but only with human oversight and appeal rights. Just 1.3% were ready to judge their silicon supervisor strictly by the results.
Apparently, the bots can have a seat at the table. DTI readers just aren’t ready to hand them the org chart or the corner office.
DTI respondents keep the autonomy dial low
The AI-boss result was not an isolated burst of technophobia. Similar boundaries appeared repeatedly when DTI asked readers how much control they would surrender to AI.
On July 29, 72.8% of respondents said an AI agent on their work PC should be limited to offering suggestions while they clicked the buttons themselves. Another 23.7% would grant access to selected folders and apps with full audit logs. Only 3.5% were comfortable giving an agent broad access with approval required only for sensitive actions.
A June 1 poll produced a similar result. After DTI covered a more self-aware version of Anthropic’s Claude, 78.3% of respondents said they still would not trust autonomous AI at work without real-world proof. Just 6.6% were ready to trust it outright, while 15.1% could be persuaded if it saved enough time and money.
The skepticism stretches outside enterprise systems. On May 20, 62.8% of respondents said they would not let Google’s Gemini Spark handle everyday tasks for them; 32.1% would delegate only low-stakes chores. And back in March, 65.4% said they would not trust Claude to run tasks on their Mac, while another 31.5% would allow it only while keeping a close watch.
That pattern is showing up elsewhere. In Stack Overflow’s May 2026 pulse survey of 1,100 developers and working professionals, agent usage had nearly doubled from the previous year to 59%. Yet 63% rarely or never allowed agents to operate entirely on autopilot, and 60% blocked them from making unapproved changes to the system.
Even the people building AI systems are keeping a hand near the emergency brake. The 2026 AI Engineering Survey from Notion, Amplify Partners, and Vercel collected 1,053 responses and found that, among teams using agents, 90% allowed them to write data. But 65% allowed agents to write data only with a human in the loop, compared with 25% granting full autonomy. Human approvals were the most common agent-control measure, used by 73% of agents.
In other words, adoption and autonomy are not the same thing.
Workers aren’t rejecting AI; they’re assigning it a role
The distinction becomes clearer when AI is positioned as an assistant rather than the final authority.
In an Aug. 13 DTI poll, readers were asked whom they would trust most to forecast a major storm. A whopping 92.4% of respondents chose humans and AI working together. Only 5.1% preferred human meteorologists alone, while 2.5% put their faith solely in the most accurate AI model.
Healthcare produced a similar, although more cautious, result. Asked in May whether they would trust AI to assist with emergency-room diagnoses, 59.7% wanted strict human oversight. Another 27.1% were comfortable using AI as clinical decision support, while only 13.2% rejected its involvement outright.
Those findings rhyme with a much larger Quinnipiac University national poll conducted in March. Among 1,397 US adults, 80% said they would be unwilling to work under an AI supervisor that assigned tasks and schedules — strikingly close to DTI’s 85.4%.
Quinnipiac also presented respondents with a hypothetical medical AI proven to be more accurate than humans at reading scans. Despite that advantage, 81% still preferred a combination of AI and a human, versus 14% choosing a human alone and just 3% choosing AI alone. The poll also exposed a broader trust gap: 76% said they trusted AI-generated information only some of the time or hardly ever.
Medical research points in the same direction. A study published in JMIR Formative Research in April 2026 found only 7.8% preferred an AI-first diagnosis that a physician subsequently reviewed. By comparison, 26.7% wanted a doctor to diagnose first and then use AI as a second opinion, while 62.8% most trusted physician expertise alone.
The study also found 81.1% were very or extremely concerned that future doctors could rely too heavily on AI without critically questioning its results. Its researchers characterized the public’s position as a form of guarded acceptance: AI may be welcome, but the human professional is expected to remain accountable.
Even weather forecasting shows the trust gap. An August Met Office study of more than 6,000 UK adults found that roughly 80% were confident using established physics-based forecasts to make decisions, compared with around 40% for potential AI-generated forecasts. The Met Office itself is exploring blended AI and physics-based approaches rather than tossing the old models out the window.
The questions and populations differ, so those figures should not be compared head-to-head with DTI’s polls. The recurring theme, though, is difficult to miss: AI becomes easier to trust when expertise, accountability, and final judgment still have a human address.
The workplace data says ’copilot’ for a reason
Research focused specifically on working professionals makes the distinction even sharper.
G2 Research interviewed 121 business professionals about AI at work and found that 80% described their organizations as supportive of AI. Eighty-two percent were most willing to hand administrative and documentation work to AI, showing a strong appetite for delegation.
But the line moved when judgment entered the picture. Forty-five percent said relationship-driven work remained fundamentally human, while 41% reserved strategic judgment and leadership for people. Seventy-five percent cited reliability and the need for human review as their primary AI concern, and 74% wanted humans to retain the final say over AI-assisted strategy and oversight.
Employees weren’t necessarily losing value in that arrangement. Sixty-eight percent told G2 that AI had increased their value at work, and 62% expected their jobs to shift from routine execution to oversight and review.
The latest CNBC and SurveyMonkey Quarterly AI and Jobs Survey similarly suggests that AI is becoming normal without becoming the default decision-maker. Roughly half of workers reported using AI at least weekly, yet the broader picture was one of uneven adoption, limited training, and unclear workplace rules.
That is less “humans versus AI” and more “humans have discovered a very fast intern who doesn’t need lunch — but still needs someone to check their math.”
Managers are experimenting the same way. Research covered by CIO Dive found 72% of managers considered public AI tools useful for preparing difficult employee conversations. That is AI acting as a management adviser rather than becoming management itself.
The governance footnote is sizable (and slightly terrifying): 44% of managers had entered employee names and performance details into those tools, while only 45% of surveyed organizations had a formal written AI policy for performance management. So, your boss might be consulting ChatGPT before your next difficult conversation, and HR may not have written a rule about it yet.
Not every finding favors keeping AI in the passenger seat
There are signs that some workers may be more comfortable with algorithmic leadership than the headline numbers suggest.
A 2026 Scientific Reports study led by Zhejiang University researchers, using China-based participants, found that participants were more willing to voice honest opinions to algorithmic leaders than human leaders across three experiments. Researchers linked the effect to perceived fairness and psychological safety. The advantage appeared for cognitive tasks involving areas such as decision-making and performance evaluation, but disappeared for emotionally oriented tasks.
That distinction matters. An AI boss may be terrible at empathy and context while still appearing less likely to play favorites around a spreadsheet.
There is another paradox: people who do not want AI as the boss may happily consult it about the boss. Resume Now’s “AI Boss Effect” research found that workers frequently turned to ChatGPT for workplace advice rather than their managers, often because the chatbot was perceived as easier to approach or less likely to judge them.
That does not necessarily show a desire for autonomous leadership. If anything, it reinforces AI’s growing role as the office confidant, coach, and second opinion.
Executives also appear willing to push further toward autonomy than many workers. Deloitte’s August 2026 survey of 501 US business and IT leaders found 61% expected that within four years, most of the AI agents their organizations use will be largely autonomous, with humans serving primarily in an oversight role.
But even Deloitte’s respondents put an asterisk on the robot takeover. Seventy-five percent said human collaboration with AI agents creates more value than agent-powered automation alone, while 70% said their organizations still could not adequately trust and govern agents. Only 15% had scaled orchestrated, multiagent deployments.
The future may be agentic. The permission dialog is apparently coming with us for the ride.
What enterprises should take from DTI’s polls
The DTI findings suggest the challenge for enterprise AI isn’t simply convincing people that the technology is useful. Plenty of respondents already appear convinced.
The harder question is how much authority an AI system should have before a person must approve, review, or reverse its actions.
That distinction has immediate implications for enterprise deployment. Agents that can act across applications should start with narrow permissions, clear audit trails, defined escalation thresholds, and explicit approval requirements for consequential actions. TechRepublic has reported that AI-agent permissions are becoming an enterprise security gap, with organizations struggling to govern agents operating under access models originally designed for humans.
Businesses should also distinguish between delegating a task and delegating accountability. Summarizing notes, drafting reports, or organizing information may need little intervention. Hiring, firing, medical decisions, financial transfers, security changes, and other high-stakes actions deserve a much higher threshold.
That is especially important because worker resistance may involve more than doubts about model accuracy. August polling from Groundwork Collaborative and Ipsos found roughly two-thirds of workers expected workplace AI adoption to make work worse through job losses or increased pressure, while 51% believed the benefits would flow mostly or entirely to owners and executives.
Deploying AI without worker input can therefore turn a technology implementation into a trust problem.
The best enterprise pitch may not be “Look how much this AI can do without you.” It may be the opposite: Look how much more you can do while remaining in control.
Across months of DTI Pulse Checks, respondents have repeatedly indicated that they are willing to put AI to work. They just want humans holding the badge, the steering wheel, and — when necessary — the big red stop button.
Methodology: DTI Pulse Checks are voluntary reader polls, not scientific surveys designed to represent DTI’s entire audience or the broader public. Poll results in this report reflect responses collected after each poll remained open and may therefore differ slightly from preliminary results published in the following day’s newsletter. Because each question drew a different respondent group, the results are presented individually rather than pooled into a single figure.