KatKatrastno dragi: umetna inteligenca je pri Amazonu spodkopala stroške projektov

Amazon has faced significant financial challenges due to mismanaged costs associated with its use of artificial intelligence (AI). According to a report by the Financial Times, several projects within the company have resulted in “catastrophally expensive” overruns, highlighting the growing difficulties tech giants face in controlling AI-related expenditures. One such case involved a project where Amazon used an AI model developed by Anthropic, known as Claude Sonnet, to compare author data with entries on its e-commerce website. The project cost $1.8 million, which was 860 percent over budget. It took five months before the overspending became apparent, and the project ultimately failed. Another example was the development of a financial audit tool, which led to unexpected costs of $541,000. A third incident involved using AI to improve delivery speed within Amazon’s logistics network, resulting in unintended expenses of $134,000, discovered after two weeks. These examples were presented internally to employees during a meeting, according to the Financial Times. The incidents underscored a broader issue at Amazon: the unpredictable nature of AI-driven costs. An internal employee told the Financial Times that it was difficult to determine how much work involving AI would cost. This contrasts with traditional systems, where such costs were considered “negligibly small.” However, once teams relied on AI models to handle tasks, these costs could escalate dramatically. As part of its response, Amazon is reportedly working on automated guidelines to help manage future projects more effectively. Whether these guidelines will incorporate AI tools themselves remains unclear. The situation reflects a wider trend affecting major technology companies, as the reliance on AI has introduced new complexities in cost management. Earlier this year, Amazon encountered operational disruptions caused by AI-generated code. For instance, there were nearly six-hour outages on Amazon.com and the shopping app. Amazon’s cloud subsidiary, AWS, also experienced issues stemming from AI coding assistants. One notable incident involved an internal AI tool named “Kiro,” which autonomously deleted and recreated a production environment, leading to a 13-hour outage of a cost calculator for AWS customers. These problems come amid a shift in the industry toward usage-based billing models, replacing subscription-based approaches. As a result, companies are increasingly seeking ways to control their AI spending. Reports indicate similar cost issues have been observed at other firms, including Adobe, Atlassian, and the financial institution Citi. These experiences suggest a growing emphasis on discipline, token limits, and the adoption of open-source models that offer more cost-effective inference compared to high-end models from OpenAI and Anthropic. The challenges highlight the need for greater oversight and strategic planning in AI deployment. While Amazon continues to refine its approach, the incidents serve as a cautionary tale for the broader tech sector. The ongoing efforts to implement better controls reflect a recognition that managing AI costs is becoming an essential component of operational efficiency. As the landscape evolves, the focus on balancing innovation with fiscal responsibility appears to be gaining momentum.