Workers once reached for a company-approved spreadsheet or email client. Now they juggle agents, chatbots, code assistants and specialized models that pop up daily. The result looks less like progress and more like digital clutter. Companies burn cash on overlapping subscriptions while employees waste hours coaxing usable output from half a dozen systems.
The Scale of the Problem
Business Insider first flagged the trend in a June 21, 2026 article that captured the exhaustion many knowledge workers feel. (Business Insider) Its survey of 6,000 digital professionals across the US, UK and Australia delivered stark numbers. Seventy-seven percent of AI users tap multiple programs each week. A full third rely on four or more. Sixty percent recycle the same prompts across tools when the first result disappoints.
Those individual time savings sound impressive on paper. Users report an average of 11 extra hours a week. Yet only 13 percent say the gains have meaningfully lifted company performance. The gap between personal productivity claims and organizational results keeps widening.
Executives have started to notice. Meta and AT&T reportedly moved to limit AI usage after costs climbed faster than expected. Amazon scrapped an internal leaderboard that rewarded raw token consumption after employees gamed the system with pointless queries. Palantir CEO Alex Karp compared the obsession to a more familiar vice. Duolingo reversed course on tying AI usage to performance reviews.
Lee Senderov, chief transformation officer at Travelport, watched one employee consume 160 times more tokens than the next highest user over four days. The pattern repeats across departments. Sales reps generate near-identical reports in isolation. Marketers produce similar decks. No one shares the refined prompts or the lessons learned from failed experiments. Hard costs multiply. So do the softer expenses of duplicated effort and lost expertise.
But. The real damage runs deeper than invoices.
Rebecca Hinds, who heads Glean’s Work AI Institute, frames the issue as a classic tragedy of the commons. Individuals grab the tool that boosts their own output. They rarely pause to consider the drag on team coordination or institutional knowledge. The result is fragmented work, reduced trust and a quiet erosion of collaboration that once happened naturally over coffee or Slack threads.
Kate Niederhoffer, head of BetterUp Labs, puts it plainly. “The pressure to signal innovation by mere AI awareness, knowledge, appetite, is so strong, and it’s leading us astray.” Companies rarely answer the foundational questions. Why adopt this particular tool? What problem does it solve? How will success look across the organization? Without those answers, adoption becomes theater.
Emily DeJeu, a professor at Carnegie Mellon’s Tepper School of Business, reaches back to economist Herbert Simon’s concept of satisficing. People settle for work that is good enough rather than optimal. Organizations exist to coordinate those individual shortcuts into something greater. Mass layoffs paired with AI rhetoric that one person can now do the work of a team threaten to break that coordination.
The Wall Street Journal explored a related phenomenon in May 2026. Companies now wrestle with “AI agent sprawl” as nontechnical employees spin up autonomous bots using platforms such as Anthropic’s Claude Cowork. (The Wall Street Journal) Walmart and others report management headaches, cybersecurity exposure and ballooning expenses. Gartner projects the average Fortune 500 will manage 150,000 AI agents by 2028, up from fewer than 15 in 2025. Many of those agents remain undiscovered, over-permissioned and lacking basic lifecycle oversight.
Security teams sound the loudest alarms. A June 10 report from Utility Dive revealed that three-quarters of organizations lack full visibility into the user accounts created for AI agents and copilots. (Utility Dive) Each new identity becomes a potential entry point for attackers. GitGuardian’s 2026 secrets sprawl analysis showed AI-service credentials leaking at record rates, with over 1.2 million exposed in 2025 alone, an 81 percent jump year over year.
Recent analysis from Atlan highlights two intertwined problems: identity sprawl and context sprawl. (Atlan) Agents accumulate without a central registry. Each builds its own narrow, often contradictory understanding of company data. The same question yields different answers depending on which bot you ask. Enterprises that once worried about shadow IT now face shadow agents operating at scale.
Airia’s May 2026 assessment warned that governance gaps, security risks and cost overruns have accelerated faster than leaders anticipated. (Airia) TaskFord’s blog post echoed the concern, noting that information scattered across dozens of systems makes delivery timelines harder to predict and control.
So companies experiment with fixes. Some create AI centers of excellence. Others build internal registries or require approval before new agents go live. Travelport tries to spot overlapping projects early and push colleagues to collaborate rather than duplicate. Success remains uneven. Larger organizations struggle most. The tools arrive faster than policies can adapt.
Critics point out that the rhetoric around AI replacement has fueled individualism at the expense of collective capability. Mark Zuckerberg once suggested a single employee could handle work that required entire teams. That message lands differently when teams feel replaced rather than augmented. Trust slips when colleagues suspect the polished slide deck in the shared drive came from a prompt rather than human insight. BetterUp’s earlier research showed that “workslop” — low-effort AI output — reduces coworkers’ willingness to rely on the person who produced it.
Yet the technology also opens doors. Non-coders now prototype simple applications. Marketers test campaign ideas without waiting for engineering support. Startups operate with leaner staffs. The challenge lies in moving those scattered wins from individual laptops into shared, repeatable processes that deliver consistent value.
Token-maxxing, the spring 2026 obsession with raw usage metrics, already feels dated. Leaders who chased volume have pivoted toward outcomes. The smarter ones ask harder questions about coordination, governance and measurable business impact. They treat AI less like a personal superpower and more like infrastructure that requires maintenance, standards and accountability.
The sprawl won’t disappear on its own. Left unchecked, it risks turning promising technology into an expensive distraction. Companies that impose structure without killing experimentation stand the best chance of converting scattered experiments into durable advantage. The rest may find themselves managing an ever-growing pile of half-used tools and disappointed expectations.
WebProNews is an iEntry Publication