Your Employees Are Using AI. But Are They Still Thinking?

Artificial intelligence has moved from novelty to daily habit inside a remarkably short window. For business owners and executives across the cleaning, facility solutions, and restoration industries, the technology now sits inside inboxes, spreadsheets, and customer service workflows whether leadership planned for it or not. The bigger question isn’t whether employees are using AI. It’s whether they’re still thinking for themselves while they do.

Matt Rosen, founder and CEO of Allata, a 10-year-old technology consulting firm with roughly 400 employees across the United States, Argentina, and India, has spent the past three and a half years helping organizations answer that question. Allata started as a custom software development shop, moved into data and early machine learning work in 2019, and went all in on AI after ChatGPT’s public launch in November 2022.

A familiar kind of disruption

Rosen pushed back on the idea that AI should keep business leaders up at night. He compared the current moment to the dot-com boom, when every job title suddenly carried an “e” in front of it, and to earlier shifts from desktop computers to the internet to mobile phones. The pattern, he argued, tends to repeat: loud predictions of mass job loss followed by hiring, not layoffs, at the companies that adapt fastest.

“Most organizations out there are, while they’re being impacted, new roles are being created,” Rosen said. He added that the businesses seeing the strongest results tend to be traditional companies using AI to strip out inefficiency and find new revenue, not just tech-forward startups.

Spotting the warning signs

Rosen described AI as a thinking partner, not a replacement for judgment. He compared it to a junior associate that’s useful for brainstorming, drafting, or tightening a report, but not for producing final work that no one on the team can explain.

His test for whether an employee is still engaged with their own work: ask them to defend it out loud. “A good way to know if employees are using AI and not thinking is asking them to actually defend and present the work that they’ve submitted, whether it be a spreadsheet, a Word document, a PowerPoint, an email,” Rosen said. “If they can’t orally cover what was in there, there’s a good chance that they’re not thinking about it.”

Direction first, then training

Banning AI outright, Rosen said, is shortsighted, since most employees already have free or low-cost access to ChatGPT, Claude, or Gemini on their phones regardless of company policy. He recommended organizations instead set clear expectations about which tools to use and, ideally, connect AI to their own data and documents in a secure environment rather than leaving employees to search the open internet or upload company data to a public access model.

Access without training, he cautioned, produces uneven results. Some employees will figure out how to multiply their output; most others won’t touch the tool or won’t use it well. Rosen pointed to a recent industry study showing roughly 93% of AI spending going toward tools and only about 7% toward training. “That’s just way too lopsided,” he said. His advice for companies unsure where to start: use AI itself to help build internal training based on the organization’s own inefficiencies and goals.

Making AI-assisted work your own

Content that reads as obviously AI-generated carries a credibility cost, Rosen said, whether it’s an email, a LinkedIn message, or a video. He described disconnecting from LinkedIn contacts who let AI handle their outreach entirely, and said the same instinct applies broadly: using AI to check or sharpen work is valuable, but passing along unedited output erodes trust.

He also flagged a cost dimension many leaders overlook. Employees often default to the most expensive, most capable model available for routine tasks like drafting an email, when a cheaper model would do the job. Allata has started setting usage guidelines internally, directing staff toward specific models for specific tasks to keep spending in check as model costs and usage scale up.

What comes next

Rosen expects model capability to keep advancing quickly, even as some providers raise prices for premium tools. He pointed to a concept he picked up from the podcast The AI Daily Brief, hosted by Nathaniel Whitmore, sometimes called the “capabilities overhang”: the idea that AI can already do far more than most organizations are currently asking of it.

Inside Allata, Rosen said the clearest gains have come in sales, where staff can synthesize client information and build proposals in a fraction of the previous time. In addition, Allata used AI to cut about $500K out of the semi-annual review process, taking the time to write a review from 8-10 hours to 30-60 minutes. Looking further out, he expects AI-driven agents to take on more routine tasks such as password resets and basic customer service work, will impact routine knowledge work, and predicted that within five to ten years, AI-powered robots will begin handling dangerous work in fields such as specialty chemicals and deep-sea operations.

“AI is not coming for your job. It doesn’t want your job. It just does what you tell it,” Rosen said. “But someone who’s using it and using it powerfully, they’re coming for your job, they might be coming for your company, they might be disrupting business models.”

The infrastructure question

Asked about local opposition to data center construction, including in Ohio, Rosen said the computing capacity behind AI has to come from somewhere, and states that find ways to power data centers, whether through nuclear, natural gas, or other sources, stand to benefit most. He argued that communities and data center operators could do more to negotiate mutual benefit, such as subsidized energy costs or local investment, rather than treating new facilities purely as something to resist.

For the business owners and managers Rosen works with day to day, his overarching message is less about prediction and more about posture: get familiar with the tools now, set clear boundaries for employees, and treat training as seriously as the technology purchase itself. This technology will only continue to advance and the time to start leveraging it is now.