Amazon and AWS to be more cautious with AI-generated code

• 
  
  Internal memo from Amazon calls staff to meeting about AI-generated code.
  
  


• 
  
  Human oversight by senior engineers necessary before production deployment.
  
  


• 
  
  Claims that job cuts exacerbate outages.
  
  

Amazon’s retail business bosses have called a meeting of engineers to examine a series of recent outages in the company’s e-commerce systems. Internal briefing material obtained by the


Financial Times


[paywall] described a pattern of incidents with operational impact. Several incidents involved software changes produced with the assistance of generative AI coding tools. Amazon’s internal messages noted the use of these tools remains new in many teams and that established safeguards are still incomplete.

Senior vice-president Dave Treadwell informed staff that site availability had fallen below the level the company expects. The weekly “The Week in Stores Tech” meeting is to therefore concentrate on an examination of recent failures and on immediate operational changes intended to reduce the risk of further incidents. Attendance at the meeting, which is normally optional, has been strongly encouraged.

The concern reflects several operational events. Earlier this month Amazon’s website and shopping application were unavailable for close to six hours, with the company attributing disruption to erroneous software deployment. Customers were unable to complete purchases and could not access routine and simple account functions like checking product prices or reviewing account details. The internal note circulated before the meeting described a trend in which incidents affected large portions of the company’s systems after software changes being produced with generative AI assistance.

The note stated that best practice for these tools remains under development – in other words, it’s early days. Generative systems can produce code quickly, but lack a full understanding of system dependencies and context in a large production environments. As a result, code changes that are seemingly correct in isolation can cause disruption once deployed in production.

Amazon intends to require additional human oversight when generative AI tools are used by developers. Junior and mid-level engineers will need approval from senior engineers before AI-assisted code changes can be released. The aim is to ensure that every change receives a review from someone with a deeper understanding of the production system, placing human judgement at the centre of development processes. The responsibility for evaluating risks from AI code will remain with engineers who understand the operational environment.

Related incidents have also occurred in the company’s cloud division, Amazon Web Services. In December, a cost calculator used by customers experienced a thirteen-hour interruption after engineers permitted the company’s Kiro AI coding tool to make changes. According to previous reports in the

FT

, the tool decided it was necessary to delete and recreate its operating environment. Amazon later described the disruption as limited in scope and affected the service in some parts of China.

Engineers at the company have recently pointed out operational pressures, with some claiming teams are handling a greater number of ‘Sev2’ incidents, which require rapid response to prevent service outages. The comments follow several rounds of corporate job cuts, including the elimination of about 16,000 roles in January, but Amazon disputes that staffing reductions have contributed to recent incidents.

The events illustrate a constraint in the use of AI coding tools. While systems may accelerate code generation, they do not replace the need for review by engineers who understand how changes interact with huge systems. Human oversight remains necessary to assess risk and prevent local code changes from producing ensuing failure. In large production environments like Amazon’s retail platform and AWS infrastructure, that review process remains essential to reliability.

(Image source: “20161005-FNS-LSC-0754” by USDAgov is marked with Public Domain Mark 1.0.)

Want to experience the full spectrum of enterprise technology innovation?

Join

TechEx

in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click

here

for more information.

TechHQ is powered by

TechForge Media

. Explore other upcoming enterprise technology events and webinars

here

.