Amazon’s AI boom is creating a new kind of disruption: internal tool bloat and data duplication.
Some teams are rapidly building their own AI-powered applications to automate workflows and organize information. But that burst of creativity has also led to problems with software and data duplication, according to internal documents obtained by Business Insider.
“AI is exacerbating the problem of duplication of tools,” the document states. “More duplicates are being created faster and fewer duplicates are being cleaned up.”
This trend signals a broader shift across corporate America. Generative AI is creating what some refer to as “AI sprawl.” This is a proliferation of AI tools and autonomous agents that threatens to overwhelm a company’s centralized monitoring and security controls.
As more employees launch tools themselves, and in some cases using AI assistants to launch tools in minutes, organizations can lose visibility into what systems are being used, where sensitive data resides, and how much redundant software they maintain.
Cloud —> SaaS —> AI sprawl
It’s a familiar story with a twist. When cloud computing arrived 20 years ago, rogue employees activated Amazon Web Services accounts without permission. Then, “SaaS sprawl” spread cloud software across enterprises, initially with little oversight.
In both cases, companies ultimately responded by taking extreme measures to govern, establish oversight, and design new systems to formally leverage these powerful new technologies. Generative AI is in the early stages of the same pattern, but is progressing much faster.
The impact of this unruly creative AI boom could be especially pronounced on Amazon. The company is implementing AI across its operations, and CEO Andy Jassy has urged employees to adopt the technology or risk falling behind.
“Drastically lowering the wall”
Amazon has been working with teams developing similar tools in parallel for years, according to documents obtained by Business Insider. These duplications were costly but manageable. Building the software required significant time and engineering resources, and maintaining the software was taxing enough that some redundant tools were eventually retired.
AI is changing that equation.
According to the document, AI will “dramatically lower the barrier to building new tools” and allow teams to prototype and ship software much faster. Instead of looking for existing solutions, engineers can create their own solutions faster and maintain them at a much lower cost.
The result is a proliferation of duplicate systems and less pressure to integrate them. “AI is currently exacerbating this problem in both directions,” the document states.
The document, marked “Amazon Confidential,” was created in February by a team of thousands of engineers responsible for evaluating and improving AI tools used across Amazon’s vast retail operations. The company encourages open dialogue about challenges, and this approach has helped it stay ahead of problems in the past.
Amazon spokesperson Montana McLachlan said in an email to Business Insider that the document reflects the perspective of a single team, and that it would be “inaccurate” to use one group’s views to characterize the experience of the company’s broader workforce.
“Artifacts remain”
The problem goes beyond tool overload. New risks are also emerging in the way data is processed and stored.
Many of Amazon’s AI systems ingest internal data and transform it into new formats, such as knowledge bases and summaries. These outputs are often stored separately from the original source, creating new copies of virtually the same information, according to the document. If the original data is later deleted or access is restricted, those derived versions will not necessarily be updated.
In one internal case, a system called Spec Studio continued to display details of software that had been kept private in Amazon’s internal code repositories, according to the documents. The company is now asking its team to better document how it handles permission changes and data deletion.
“Any system that ingests data, transforms it through AI, and stores the output separately faces the same problem: When source permissions change or data is deleted, derived artifacts remain,” the document states.
Use AI to solve AI problems
The rapid proliferation of AI-generating tools is creating “shadow AI” within organizations, and rogue applications pose risks such as leaking sensitive data and violating regulations, Debo Dutta, chief AI officer at cloud company Nutanix, told Business Insider.
“If not managed properly, all this can lead to data and system disruption,” Dutta said.
Amazon’s answer may be AI.
The company is exploring ways to use AI to identify duplicate tools, flag risks and encourage teams to integrate early, before duplication becomes difficult to resolve, the document said.
The challenge is to balance speed and coordination. Amazon has long valued the autonomy of small, independent teams to move quickly and make their own decisions, through an approach often referred to as the “two-pizza team” model. That culture encouraged rapid experimentation, but it could also make problem solving difficult.
“Teams building bespoke AI systems are likely to repeat similar problems,” the document states.
Any tips? To contact this reporter via email, please specify the following address: ekim@businessinsider.com or on Signal, Telegram or WhatsApp 650-942-3061. Use a personal email address, non-work WiFi network, and non-work device. Here’s a guide to sharing your information securely.