There’s a recent story that should make any business owner or operations director think twice if they’re pushing their team to “adopt AI as soon as possible.” According to a Fast Company report, Amazon employees are under so much pressure to increase their use of AI that they’ve started making up tasks just to hit the metric. Yes, you read that right: skilled people wasting time pretending to use AI instead of using it for something useful.
This is not an isolated case or a random anecdote. It’s a symptom of a management mistake that is being repeated in thousands of companies right now. And it’s worth unpacking, because the core lesson is the difference between adopting AI well and turning it into another layer of bureaucracy disguised as innovation.
What’s actually happening at Amazon
The pattern is easy to understand:
- Leadership decides AI is a strategic priority.
- Internal metrics are imposed: use this tool, this many times per week, in this many processes.
- Employees, with no clear training or real use cases adapted to their work, face a dilemma: meet the metric or do their job well.
- Many choose the most rational option from their point of view: manufacture artificial usage. They ask the AI for things they don’t need, generate summaries nobody will read, automate tasks that were already solved.
The result is the opposite of what leadership wanted: less productivity, more friction, and growing distrust of AI inside the company itself.
When you force someone to use a tool without understanding their work, you’re not measuring AI adoption. You’re measuring performative compliance.
Why this happens (and not just at Amazon)
This phenomenon has a name and a long history: goal displacement. When the metric replaces the real objective, people optimize the metric. We’ve seen it with calls per hour in call centers, with “activity” KPIs in sales, and now we’re seeing it with AI.
The causes are usually threefold:
| Cause | What happens in practice |
|---|---|
| Top-down imposition without diagnosis | Leadership decides “we’re going to use AI” without knowing the real problems each team faces. |
| Generic tool for specific problems | A license for a general-purpose assistant is bought and everyone is expected to figure out how to apply it to their role. |
| Usage metrics, not outcome metrics | The company measures how often it’s used, not how much time it saves or what problem it solves. |
The underlying mistake is treating AI as if it were Office: a horizontal tool everyone learns on their own. It isn’t. AI applied to business works when it solves a concrete problem in a concrete process, with a concrete workflow.
The approach that actually works: start with the problem, not the tool
The right question is not “How do we get the team to use more AI?” The right question is “Which repetitive tasks are consuming my team’s time, and which ones can a machine do better?”
That’s a major shift in mindset. Instead of pushing technology down, you listen from the ground up to what’s actually hurting. And then you design a specific solution.
At Studio SmartWork, we approach every client this way:
- Step 1 — Real-time time audit. Where is the team’s time going? Which tasks are repetitive, predictable, and rule-based? Those are the obvious candidates.
- Step 2 — Custom design. We don’t give the team “another tool.” We build a workflow that runs automatically in the background. The team doesn’t have to learn anything new: the solution runs and delivers results.
- Step 3 — Measure by outcome. We don’t measure “how many times you used AI.” We measure: how many hours did the team get back? How much did response time drop? How many errors disappeared?
The difference is huge. In the Amazon model, the employee has to fit AI into their day. In the right model, AI fits into the process and frees the employee.
Signs your company is falling into the same trap
If you recognize three or more of these signs, it may be time to rethink the approach:
- Your team gets trained on AI tools, but nobody knows exactly what problems they’re supposed to solve with them.
- You measure adoption by counting logins, prompts, or active subscriptions.
- You’ve bought AI licenses for the whole team “just in case.”
- Employees share “use cases” that feel forced or trivial.
- No one in the company can tell you, in euros or hours, how much AI has saved over the last 3 months.
- The AI conversation in meetings revolves around which tool to use, not which problem to solve.
What you should do instead of “pushing more AI”
This is where things get practical. If you’re leading a company and you want AI to deliver real results, this is the right order:
1. Map friction, not opportunities. Ask the team what tasks they hate, what wastes their time, and where the process breaks. That’s where the gold is.
2. Prioritize by impact and frequency. A task done 50 times a day at 10 minutes each is a better candidate than one done once a month for 3 hours.
3. Build specific solutions, not generic ones. A voice agent that handles calls with your business information. A system that qualifies leads using your sales team’s exact criteria. An inbox organized according to the real priorities of your operation. That is applied AI. The other thing is theater.
4. Measure results, not activity. Hours recovered, response times, conversion rate, errors eliminated. If it can’t be measured in one of those units, it’s not a good automation.
5. Maintain and improve. An AI solution is not a project you deliver and forget. It needs monitoring, adjustments, and continuous improvement as the business changes.
The pressure paradox
There’s something almost ironic about what’s happening at Amazon: the more you pressure people to use AI, the more likely they are to use it badly. Pressure creates compliance, not creativity. And applied AI for business requires the second, not the first.
The companies getting real value from AI are not the ones with the most aggressive adoption metrics. They’re the ones that have identified two or three critical processes and built surgical solutions for those processes. Less noise, more results.
That’s what we call smart work: the machine does machine work, and people focus on human work — thinking, deciding, creating, and building relationships with customers.
Conclusion: AI isn’t adopted, it’s designed
The Amazon case is a useful warning. Not because Amazon is clumsy — it’s one of the most sophisticated companies in the world — but because it shows that even they are not immune to the basic mistake: confusing usage with value.
If you’re thinking about introducing AI into your business, forget the question “Which tool should we buy?” The right question, the only one that matters, is: “What specific problem do I want to solve, and how do I measure that it’s solved?”
From there, everything else is execution. And execution, when it’s done well and tailored, doesn’t take months. It takes days.
That’s what separates AI that works from AI that is theater.