Why Giving AI To Every Employee Backfires

Why Giving AI To Every Employee Backfires

Handing an AI subscription to every employee sounds like progress until you look at what actually happens to your output. Spoiler alert: it rarely makes weak habits better.

Most companies treat AI access like free coffee in the breakroom. Just turn it on, tell people to "be innovative," and wait for magic. Yet at global payments firm Paysend, Group CEO Ben Chisell took the opposite route, restricting general-purpose AI tools to proven high performers and requiring staff to show real value rather than chasing hype. That move goes straight against the grain of blanket rollouts like Lloyds Banking Group's deployment of 40,000 Microsoft 365 Copilot licenses or Grant Thornton UK's push for staff-wide Claude usage.

If you've watched teams burn cash on prompts that produce mediocre work faster, Chisell's skepticism won't surprise you. AI doesn't fix fuzzy thinking or weak judgment. Often, it just speeds up the wrong direction.

Why Restricting AI Access Makes Sense In Practice

Let's be honest about what happens when you give everyone unrestricted access to generative tools on day one. People who already struggle to scope a problem, verify facts, or spot bad logic can suddenly generate five times as much flawed work before lunch.

Chisell has pointed out that AI can amplify bad habits like laziness and poor judgment when handed out without proof or guardrails. A former product leader at Amazon, eBay, and Starling Bank, he frames his approach around business outcomes and operational discipline rather than "tokenmaxxing" or fear of missing out. When you test AI with high performers first, you see what the tool actually moves—and you build evidence before asking the rest of the company to adopt it.

Even outside Paysend, other leaders are starting to pump the brakes on unchecked access. Reports in late 2026 highlighted firms like Goldman Sachs, JPMorgan Chase, Uber, and Cisco tightening controls or metering usage around costs, security, and accountability. Meanwhile, research from Salt Security in November 2026 found that 77% of surveyed organizations had already seen an AI system or agent act outside intended or authorized behavior, prompting 64% of U.S. and U.K. executives to restrict, delay, or cancel an AI project over security and governance concerns.

There's a big difference between banning AI and earning the right to scale it. Treating AI as a privilege tied to clear use cases forces teams to ask a basic question: Does this actually help our customers or hit a real metric?

The Illusion Of Productivity That Trips Teams Up

Ask anyone who's spent a week prompting models all day—it feels busy. You're typing fast, getting instant drafts, and shipping code or copy quicker than usual.

That feeling is dangerous. CoreView CEO Simon Azzopardi warned that AI-generated work isn't automatically productive when employees use it outside their real expertise, noting that people can feel more efficient while actually getting less real value done. You spend hours polishing AI drafts, debugging subtle errors, or reviewing bloat that nobody asked for.

Numbers from wider research back up that gap between activity and actual business impact. U.K. government research cited in coverage of this debate found that while three-quarters of businesses using AI reported better workforce productivity, 77% saw no change in revenue. Gallup data from October 2026 also showed that U.S. employees using AI mainly report gains in speed (63%) and creativity (56%), with regular use concentrated heavily among managers and more educated workers.

  • Speed isn't outcome: Cranking out ten mediocre options doesn't beat picking one sharp solution.
  • Expertise still matters: Without domain judgment, it's tough to tell when a model sounds confident while being subtly wrong.
  • Hidden coordination costs: When everyone relies on personal AI shortcuts instead of talking to teammates, shared context and trust can slip, as highlighted in Harvard Business Review coverage on team dynamics.

If your team measures success by prompt volume or slide count, you're optimizing for theater. Real leverage shows up in cycle time, error rates, revenue, or customer retention.

What Selective Rollouts Get Right And Where They Can Go Wrong

Gating AI behind high performers isn't a silver bullet, and it comes with real trade-offs. If you only give tools to top staff or senior folks, you risk starving junior employees of learning reps—or creating an awkward two-tier culture where only a few people know how to work with modern workflows.

There's also a timing trap. Some leaders restrict AI out of caution, while others restrict it out of penny-pinching. Capping usage tightly can backfire if engineers or analysts hit arbitrary walls right in the middle of solving a tough problem, forcing awkward workarounds.

That's why the proof-of-value model matters more than pure elitism. At Paysend, new hires have been given a window—around a month—to test AI tools and show what they actually add. In one example reported in October 2026, a new SEO lead tested Claude for a month and concluded they didn't need a general chat tool ongoing, proposing a focused content-brief agent instead. Paysend also estimates it gets roughly 25% more engineering output with the same team size while keeping a disciplined, selective approach.

When a company ties access to documented wins rather than status alone, the goal isn't hoarding technology. The goal is turning one person's smart experiment into a repeatable workflow that helps everyone else without creating noise.

How To Test AI Without Wasting Money Or Trust

You don't need a dramatic manifesto to stop burning budget on aimless AI rollouts. Start small, stay honest about what's working, and judge every tool by how it changes actual work.

  1. Pick one measurable bottleneck first. Don't start with "let's use AI in marketing." Start with cutting QA review time, speeding up treasury reconciliation, or drafting cleaner support escalations—and define what success looks like before you buy seats.
  2. Run a time-boxed pilot with sharp operators. Put the tool in the hands of people who already have strong judgment and clear standards. Give them a set period (say, 30 days) to test real tasks and report back on what saved real time versus what created extra cleanup.
  3. Require proof before expanding licenses. If a tool doesn't improve quality, speed, or cost on concrete workflows, cut it. When a workflow does prove out, document the exact process so others can reuse it instead of reinventing prompts from scratch.
  4. Keep sensitive data locked down. Before anyone pastes proprietary code, customer records, or financial data into an external tool, make sure permissions, data-retention rules, and human sign-off steps are clear. Research from KPMG has repeatedly flagged unauthorized or risky AI data habits at work, so guardrails aren't optional.

At the end of the day, AI is an amplifier, not a substitute for clear thinking. If your team has strong judgment, focused goals, and honest feedback loops, selective adoption will make you sharper—and save you a whole lot of expensive noise along the way.

DK

Dylan King

Driven by a commitment to quality journalism, Dylan King delivers well-researched, balanced reporting on today's most pressing topics.