Many organizations are discovering that moving fast on AI without enough structure creates more problems than it solves.

Several issues are pushing companies to slow down and rethink their AI strategies:

• Quality problems and “AI slop”: Employers report a flood of low-quality, error-prone AI output. At one large tech firm, a senior executive described getting clearly AI-generated documents, summarizing them with AI again, and ending up with “no new thinking” – just an “ocean of sameness”.
• Hidden costs instead of productivity gains: Workers are spending significant time cleaning up AI-generated mistakes. This extra rework is sapping employee time, pushing up costs, and eroding the hoped-for efficiency gains.
• Security and privacy concerns: Companies like Palantir, Booz Allen Hamilton, and Nvidia are limiting employee use of certain tools because of concerns about exposing sensitive organizational data to external AI providers.
• Operational disruption and incidents: An Ernst & Young survey of companies with at least $1bn in annual revenue found that 36% of senior AI executives reported an AI incident that led to issues such as data loss, financial damage, brand harm, or operational disruption.

The result: firms that started “fast and furious” on AI are now introducing more guardrails, bringing back experienced staff into key workflows, and investing in deeper training rather than just pushing more tools.

So far, broad productivity gains from AI have been limited and inconsistent.

Some key data points from recent research and surveys:

• An Atlanta Federal Reserve working paper found that 90% of executives say their AI efforts have not yet translated into productivity gains for their companies.
• A University of Pittsburgh study published in May concluded that productivity did not rise after companies invested in AI and cut staff.
• More than 112,000 US job-cut announcements in 2026 have mentioned AI as a reason for layoffs, but stock prices tend to dip after these AI-related layoff announcements, suggesting markets are skeptical about near-term performance benefits.

Several factors are holding back performance improvements:

• Error propagation: AI often produces plausible but flawed output. Because the weaknesses aren’t always obvious, mistakes can flow through entire workflows before anyone notices, causing bigger downstream issues.
• Broken processes, just faster: As Microsoft observed, “adding agents to a broken process still leaves a broken process”. Speeding up one step in a flawed workflow often just creates a longer queue at the next.
• Misaligned expectations: Leaders are under pressure to “find productivity gains” without compromising quality, while many employees feel they’ve been told to use AI without clear guidance on where it truly adds value.

In practice, the organizations seeing more benefit are the ones treating AI as a way to reimagine workflows and skills development, not just as a quick way to cut headcount or automate existing steps.

Frontline workers and managers are feeling both the pressure to use AI and the pain when it’s not implemented thoughtfully.

Some common challenges highlighted in the source material:

• Unrealistic standards and false alerts: At Walmart, AI tools that assign tasks, set schedules, and monitor performance have produced unreasonable expectations and a stream of errors and false alerts. Examples include:
  • Nonsensical routes for fulfillment workers that make orders harder to complete quickly.
  • “Useless” safety alerts that don’t match on-the-ground reality.
  • Schedules that assume workers will perform perfectly all the time, forcing people to cut corners just to keep up.
• Training and expertise gaps:
  • More than half of employees say they are not getting the AI training they need.
  • Only 45% of managers feel equipped to effectively lead “AI natives”, even though many believe their teams are more skilled with AI than they are.
  • Workers report being told to “use these tools” without guidance on where they fit best or how to avoid errors.
• Misuse and credibility risks: When AI is used without clear standards, it can undermine trust. In one legal case, an expert relied heavily on ChatGPT to assemble a 14,500-page report in hours. The jury ultimately ordered the company to pay $61.5m, and jurors expressed disbelief that an expert would make claims “without any foundation”.

In response, companies are starting to:

• Shift from one-off, generic AI webinars to sustained, role-specific training.
• Pair early-career employees with seasoned staff so that judgment and experience shape how AI is used.
• Put clearer boundaries around where AI can be used (for example, for formatting and drafting vs. forming core expert opinions).

The pattern is clear: to get real value, organizations need to rethink how people, processes, and AI work together, rather than expecting tools alone to solve systemic issues.