Companies are learning the hard way that deploying AI tools too quickly and forcing or strongly encouraging employees to use them can have significant backfires.
The latest one seems to be Amazon, but it’s debatable whether it’s taking away the right lessons. On Tuesday, financial times The e-commerce giant reportedly convened a large group of engineers for a meeting to address recent obstacles plaguing its online retail business, including those related to its AI coding tools.
The company said in its conference briefing notes that “incident trends” are characterized by “wider blast radius” and “modifications due to Gen-AI.” The memo cites “new uses of GenAI for which best practices and safeguards have not yet been fully established” as a “contributing factor.”
“As you are probably aware, the availability of our site and related infrastructure has been poor lately,” Dave Treadwell, Amazon’s senior vice president of e-commerce services, told employees in an email. F.T..
The meeting comes after a roughly six-hour outage last week when Amazon’s shopping website and app went down, leaving customers unable to place orders. The company later blamed failures on “software code deployment.”
Another series of incidents at Amazon Web Services, the company’s cloud computing division, resulted in two separate outages after engineers allowed disastrous changes to the company’s AI coding tools. F.T. The news came to light last month. In one case, an AI tool deleted and recreated the entire coding environment.
In response to previous reports, Amazon appears to be holding its own, framing these failures as issues related to the company’s protocols around AI use and “user access control,” rather than issues with AI autonomy. The company has no intention of backing down from AI adoption, instead insisting on tightening guardrails and increasing oversight of how AI is used.
Treadwell said at the meeting that junior and mid-level engineers will request senior engineers to approve AI-assisted changes. F.T.This is a report by Mr. Treadwell asked his staff to attend meetings, which are generally voluntary.
There is no question that when AI tools need to be used, they should be closely supervised, especially in programming roles. Like other generative AI models, AI coding tools often make errors and instructions can be difficult to follow. This means that users may perform unintended actions.
But Amazon is renewing its focus on implementing more human oversight, as it aims to lay off hundreds of employees from its cloud computing division and 30,000 employees overall. Meanwhile, management is relying on programmers to make heavy use of AI tools, something employees have previously told us. F.T. The company says 80% of developers have set a goal to use AI for coding tasks at least once a week.
In summary: More AI, more coding, more human oversight, but fewer humans. Let’s see how it goes.
Learn more about AI: Insiders fear the government will nationalize the AI industry
