Somewhere in the grand tradition of employees working around pointless management mandates, Amazon workers have achieved something truly beautiful: they're using AI to fake using AI.
According to a report by the Financial Times (picked up by Fast Company), Amazon has been pushing its employees harder and harder to weave AI into their daily workflows. The pressure is real. The actual guidance on how to do that? Much less so.
Nature abhors a vacuum, and so does a KPI
When you tell people to use more AI but don't define what "more" looks like, you get exactly what Amazon apparently got - employees firing up the company's internal AI tool, MeshClaw, and creating entirely pointless AI agents just to bump up their activity numbers. Not to get stuff done faster. Not to automate anything meaningful. Just to make the metrics look good.
It's the corporate equivalent of running the treadmill while sitting on the couch next to it. The numbers go up. Nothing changes. Everyone nods approvingly.
This is what a misaligned incentive looks like in the wild
To be clear, this isn't really about lazy workers. This is a textbook case of what happens when leadership confuses activity with output. If the goal is "more AI usage" rather than "better results through AI," then rational humans will optimize for the thing they're being measured on. That's not gaming the system - that's the system working exactly as designed, just not in the way anyone intended.
The irony is rich enough to spread on toast. Amazon, one of the most data-obsessed companies on the planet, apparently set an AI adoption target without properly defining what success should look like. The result is wasted compute, wasted time, and a bunch of AI agents running around doing absolutely nothing useful.
So what now?
There's a real lesson buried under all this absurdity. AI adoption mandates without clear use cases don't drive innovation - they drive compliance theater. And compliance theater is expensive, especially when the props are AI agents spinning their wheels on make-believe tasks.
If Amazon (or any company, frankly) wants meaningful AI integration, the conversation probably needs to start with "here's a problem worth solving" rather than "here's a usage quota, good luck."
Until then, the MeshClaw agents will keep busy doing absolutely nothing, and the metrics will look fantastic.
