For many teams, AI has moved from pilot to production in everyday use. That shift has reset the bar for productivity as teams push toward AI-first workflows. However, the tradeoff is often invisible: the human work required to verify, edit and step in when the tool gets it wrong, a reality of human-in-the-loop AI with real guardrails.
A new Connext Global survey suggests the reality is more complicated. AI may speed up first drafts, but reliability still depends on humans who review, correct and take responsibility when output misses context, introduces errors or creates downstream impact.
“AI can be a powerful accelerator, but this research shows most teams are still doing the hard part, making output accurate, complete and ready for real-world use,” says Tim Mobley, president and CEO of Connext Global. “The opportunity is not just adopting AI, it is building the oversight habits that keep quality high while speed increases.”
Key takeaways:
· When respondents describe what “reliable” AI looks like, only 17% say it can run on its own. Seven in 10 (70%) define reliability as a hybrid model – AI plus light review (35%) or AI plus dedicated oversight (35%).
· This also helps explain why expectations are shifting toward more oversight, not less. Nearly two-thirds (64%) expect the need for human review or checking to increase, including 26% who expect a significant increase. As AI spreads into higher-stakes workflows the perceived need for validation rises alongside usage.
· Day to day, most users say AI requires active supervision. AI needs attention almost every time (28%) or sometimes (54%). Just 4% say it can usually run without much attention.
· Follow-up work is nearly universal, with only 4% saying they rarely do it. The most common “AI aftermath” is editing or fixing (42%) and review or approval (34%).
· Only 37% say AI is right without fixes most of the time. Nearly two in three (63%) say it is right only sometimes or less, including 45% sometimes, 16% rarely and 2% almost never.
· That gap has direct implications for productivity. When AI needs fixes, nearly half (46%) say fixing takes about the same time as doing the work manually and 11% say it takes more time. In other words, 57% report that once correction is required, the time advantage can disappear, reshaping ROI for everyday tasks.
· When respondents described where AI breaks down, the top issue was missing context – 42% say AI left out important details or context. Other common issues include causing extra work to fix or redo (32%) and sounding confident but being wrong (31%).
· About one in five (19%) say AI made a customer situation worse.
· A majority (60%) say they have personally been involved in AI negatively affecting outcomes. That includes 18% who say it led to frustration or complaints and 11% who say it contributed to lost revenue or churn.
