Generative AI is transforming how companies build software and manage data—and according to a leaked internal document from Amazon, it's also creating a mess of duplicate tools and data silos that's becoming a governance nightmare.
Across Amazon's retail and tech divisions, more teams are using AI to quickly spin up internal applications for automating workflows, organizing information, and boosting development efficiency. The barrier to building these tools has dropped dramatically, letting engineers create systems in hours that once required weeks of resources.
But that low-cost creation power is breaking down the organizational guardrails that used to keep redundancy in check. When building and maintaining software was expensive, overlapping tools would eventually get merged or retired. Now, with AI accelerating development, duplicate tools are proliferating faster than they can be consolidated.
The document describes this trend as "AI sprawl"—a flood of similar or redundant AI applications and agent systems, without a corresponding boost in centralized management or visibility. It's not just wasteful; it makes it nearly impossible for the company to track what's actually being used or where data is flowing.
Data gets messy, fast
The data side is just as problematic. Many AI systems transform internal information into summaries, knowledge bases, or structured outputs, then store them separately. While that improves usability, it also spawns endless "data copies." When the original data is deleted or its permissions change, these derivative copies often don't update, creating inconsistencies and risks.
The report cites examples where information that should have been restricted still appears in old versions within internal systems. It highlights a fundamental flaw in how AI systems handle data lifecycle management: once information gets processed and re-stored multiple times, traceability and consistency plummet.
The rise of "shadow AI"
As employees gain more freedom to build with AI tools, "shadow AI" systems—unapproved AI applications that spread rapidly across teams—are popping up everywhere. They might boost local productivity, but they also introduce security and compliance risks, from exposing sensitive data to losing control over system governance.
Amazon is trying to fight AI complexity with more AI. The company is experimenting with automated systems to identify duplicate tools, suggest consolidation opportunities, and monitor for potential risks—essentially using the technology to clean up its own mess.
But the internal document points out that the root cause is tied to Amazon's long-standing culture of small, autonomous teams. In a highly decentralized structure, AI has supercharged the "every team for itself" mentality, driving up the cost of coordinating the whole system. This dynamic isn't unique to Amazon; it's a challenge facing any large organization embracing generative AI, as companies like Anthropic expand their security tools and OpenAI shifts toward enterprise solutions.