Amazon Grapples With Internal AI Tool Duplication Problem

Amazon is grappling with a growing internal mess of its own making. An internal document obtained by Business Insider reveals that the company's aggressive push into AI is producing a surge of duplicate tools, orphaned data, and overlapping systems—a problem the document bluntly describes as getting "worse from both directions."

The document, marked "Amazon confidential" and produced in February by a team overseeing AI tools across Amazon's retail business, puts it plainly: "AI is making our tool duplication problem worse. More duplication is being created faster, and less of it is being cleaned up."

The faster teams build with AI, the messier Amazon's internal infrastructure gets. The core issue is speed. AI has dramatically lowered the barrier to building new software—engineers can now spin up a working tool in minutes rather than weeks. The upside is faster experimentation. The downside, according to the document, is that teams are no longer bothering to look for existing solutions before building their own. In the past, the cost and effort of building software kept duplication somewhat in check. Redundant tools were expensive to maintain and eventually got retired. Now, with AI doing the heavy lifting, those natural brakes are gone.

The problem compounds when you factor in Amazon's famously decentralized structure. The company's "two-pizza team" model—small, autonomous teams empowered to move fast—has long been a competitive advantage. But that same independence makes it harder to coordinate, consolidate, or even track what's being built across thousands of engineers.

Private data, public problem: how Amazon's AI systems are holding onto what they shouldn't. It's not just tool overload. The document flags a trickier issue around data. Many of Amazon's AI systems ingest internal information and transform it—producing summaries, knowledge bases, and other derived outputs that get stored separately. When the original source is later deleted or made private, those copies don't always follow.

Business Insider reported one specific example: a system called Spec Studio kept surfacing software details that had already been made private in Amazon's internal code repository. The document warns that "derived artifacts persist" long after the source data has been restricted or removed.

Amazon's proposed fix, perhaps fittingly, is more AI—tools that can detect duplication, flag risks, and push teams toward consolidation before the overlap becomes irreversible. Amazon told Business Insider the document reflects the view of a single team and shouldn't be taken as representative of the company's broader experience.