When Dina, a software developer based in New York, joined Amazon two years ago, her job was to write code. Now, it’s mostly fixing what artificial intelligence breaks.
The internal AI tool she’s expected to use, called Kiro, frequently hallucinates and generates flawed code, she says. Then she has to dig through and correct the sloppy code it creates, or just revert all changes and start again. She says it feels like “trying to AI my way out of a problem that AI caused”.
“I and many of my colleagues don’t feel that it actually makes us that much faster,” Dina said. “But from management, we are certainly getting messaging that we have to go faster, this will make us go faster, and that speed is the number one priority.”
Just days after speaking to the Guardian, Dina was laid off.
Lisa, a supply chain engineer who has worked at Amazon for over a decade, says that AI tools at work have been helpful to her only in about one in every three attempts. And even then, she often finds issues and has to consult with colleagues to verify and correct their results, which takes up more time than if she’s done the task without AI.
She doesn’t take issue with the AI tools themselves, but rather the company’s logic in pushing all employees to use them daily. “You don’t look at the problem and go, ‘How do I use this hammer I have?’ she said. “You look at it and go, ‘Is this a problem for a hammer or something else?’”
More than a half a dozen current and former Amazon corporate employees, in roles ranging from software engineer to user experience researcher to data analyst, told the Guardian that Amazon is pressing employees to integrate AI across all aspects of their work, even though these workers say this push is hurting productivity. They say Amazon is rolling out AI use in a haphazard way while also tracking their AI use, and they’re worried the company is essentially using them to train their eventual bot replacements. All of this, they said, is demoralizing. The Guardian granted these workers anonymity because of their fear of professional repercussions.
“We have hundreds of thousands of corporate employees in a wide range of roles across many different businesses, each of which is using AI in different ways to learn about what works best for their use cases,” Montana MacLachlan, an Amazon spokesperson, said. “While different employees may have different experiences, what we hear from the vast majority of our teams is that they’re getting a lot of value out of the AI tools that they use day-to-day.”
This pressure comes as Amazon has laid off 30,000 workers in the last four months – nearly 10% of its roughly 350,000 corporate workforce. Its cuts are part of a wave of recent AI-connected tech layoffs, including at Block, Pinterest and Autodesk. Exactly how much these companies will be able to rely on AI to replace headcount is unclear, and each company has given an array of sometimes contradictory reasons for reductions. Jack Dorsey, the Block CEO, said outright that AI was behind his 40% staffing cuts, while Pinterest and Autodesk said they were redirecting investments to AI. Amazon has waffled in explaining how AI factors into its layoff decisions, saying both that it would lead to reductions, but that recent cuts weren’t AI-driven. The company said in February it would spend some $200bn this year on AI infrastructure and announced a $50bn investment in OpenAI.
In a moment of rising anxiety about AI and work, the decisions Amazon makes around automation – and even how it talks about these shifts – will be consequential for not just its massive workforce, but for people in industries around the world. Amazon is the second-largest employer in the US and has long influenced workplace practices across both white collar and blue collar industries.
“There’s a lot of talk among corporate employees about how some of these practices – about performance, surveillance and monitoring – are somewhat imported from the warehouse and the drivers space, and that it is Amazon expanding this model of labor to white collar workers,” Jack, a software engineer at Amazon for more than a decade, said. “It does feel like we’re at the vanguard of a new stage in employer relations with the advent of AI.”
“Half-baked” tools and “more work for everyone”
While Amazon has a reputation for being a tough place to work, the impact of its AI campaign has pressurized its workplace, workers said. “It’s worse now,” said Denny, a software engineer, who works in the retail space at the company. “If we don’t pivot ... then we risk becoming obsolete and being let go in the next layoff.”
Whenever there’s a task at hand, the biggest question managers ask is whether it can be done faster with AI tools, according to Denny. This is leading employees to use AI tools just for the sake of it. Recently, someone in Denny’s team shared that an internal AI agent had saved him about a week of developer effort on a feature. But when Denny looked at the actual code review, he found dozens of comments from colleagues pointing out basic issues. The AI generated code was full of slop.
“In the end, my guess is that the developer cycle is not going to change, and [could] even be potentially longer,” said Denny. “This pressure to use [AI] has resulted in worse quality code, but also just more work for everyone.”
Denny was one of several workers who told the Guardian they’re pressured to use an overwhelming array of AI tools, many of which were hastily developed in internal hackathons and then have to spend time answering surveys about their experience with the tools.
“I would get shown these random tools by my manager who’d be like: ‘Why don’t you try using this thing?’, and it was just the result of a hackathon,” said Denny. He says the tools are “half-baked” and unhelpful, and in fact add to his workload because he has to vet them.
Amazon typically organizes quarterly hackathons to encourage engineers to develop new projects. Sometime last year, Denny recalls, the company primarily switched to generative AI hackathons, during which the majority of projects ended up being developer productivity focused tools.
“We don’t mandate teams use AI tools,” said Amazon’s MacLachlan. “However, we believe these tools can help employees work more efficiently and automate time-consuming, undifferentiated tasks.”
There have also been public slip-ups that seem connected to Amazon’s embrace of AI. According to a February FT report, Amazon recently experienced at least two outages because of issues with the company’s internal AI tools, including a 13-hour interruption to a customer-facing system in December after some engineers allowed its AI tool “to make certain changes”. Amazon, however, said that an employee, rather than AI, caused the service interruption. The FT reported on Tuesday that Amazon would convene engineers to explore “a spate of outages, including incidents tied to the use of AI coding tools”.
“I think if you continue to push people to use AI tools in every single aspect, you’re going to get more errors like that,” Sarah, an Amazon software engineer, said.
Sarah said that AI can be useful, but its potential is best realized when engineers decide how to use it. But at Amazon, even when AI is not suited for a task, she’s now expected to train it. “We have to write out detailed procedures so that the AI can understand it and give better output,” said Sarah. “Part of my new job role, it feels like, is being asked to train the AI to essentially replace you.” She’s early in her career and worries that offloading her work to AI is stunting her learning curve.
Forcing employees to adopt tools, according to Ifeoma Ajunwa, founding director of the AI and Future of Work Program at Emory University and the author of The Quantified Worker, usually backfires. “Generally, employees are in a better position [than management] to determine what tools can aid productivity,” she said.
Meanwhile, Amazon workers are often having to seek out training for AI best practices on their own.
Will, a user experience researcher, said Amazon offers employees plenty of AI training videos on their learning portals, though most of them are optional. When he’s attended training sessions, “the focus is always, ‘here’s how to build something as quickly as possible’”. He said trainers – who are typically peer employees who are also AI power users – advise to carefully review each step before letting AI start building. At the same time, Will said: “I have been in several trainings where the instructor says you can just ask the AI to check its own work.” However, you can’t fully rely on AI to detect its own mistakes; that’s something human judgment is better suited for.
“One of the biggest predictors of AI adoption and whether employees feel that AI increases their productivity is whether management encourages it and provides training,” Alex Imas, professor of behavioural science and economics at Chicago Booth, said.
MacLachlan said Amazon provides different training and resources for people across the company, including structured options. “Employees are encouraged to use the tools themselves as a learning mechanism, adopting a learn-as-you-work approach that is proving to be one of the most practical and effective methods of AI adoption across the company,” she said.
An AI-fueled shift to surveillance
Along with the productivity challenges that have come with Amazon’s AI push, workers said it’s also making them feel surveilled.
For years, each morning when Amazon employees logged in to work, an internal system called Amazon Connections would greet them with a message and ask for feedback on topics like how their teams were functioning, or how satisfied they felt with their work. Over the last year, these questions have increasingly centered less on human factors and more on AI.
Maria, a former product manager who was laid off from Amazon in January, said questions asking her about her career or team shifted to ones such as: “‘Are you using AI in your daily work?,’ ‘How often are you using it?,’ ‘Do you think that you’re a power user?,’ or ‘Is AI a priority in your organization?’”.
Then there are more obvious indicators of surveillance. Workers said managers at Amazon have a dashboard where they track their team members’ AI use, including if they’re using certain tools and how often they do so. (The Information first reported this in February.)
Jack, the software developer who’s worked at Amazon for more than a decade, said the company also launched a different dashboard, which the Guardian has viewed, so teams could see their generative AI adoption, engagement and depth of usage. “Every team treats it differently,” he said, with some managers using it with a goal of getting at least 80% of their team using AI tools weekly.
Sarah said her team’s principal engineer told her and his other reports he checks this dashboard daily. “He’s really been pushing our AI usage,” she said.
“Of course we want to understand what tools our teams are using and whether those tools are working well for them or could be improved,” said MacLachlan.
The inevitable result of AI tools getting deployed at scale is surveillance, according to Nick Srnicek, author of Platform Capitalism and a senior lecturer in digital economy at King’s College London. “The rushed deployment of AI means an uncritical expansion of surveillance since these tools increasingly require detailed knowledge of personal workflows and data,” he said. “To make them more capable means giving management greater insight and control over workers’ everyday activities.”
Workers also said they suspect their career advancement is increasingly dependent on their enthusiastic embrace of AI.
“We have promotion documents which have a template with questions like, ‘What has this person done?’, ‘What impact did it have?’ – and now it also has a question asking, ‘How [did] they leverage AI?’,” said Lisa. “I think they want to only keep the people who support this investment [in AI] and are going to try and filter out people who do not support it or have concerns about it.” The Wall Street Journal reported in late February that at Amazon, “managers do consider who is all-in on AI when it comes to promotions”.
“While we expect employees to use resources – including AI – to make work more engaging and improve customers’ lives, we don’t instruct managers to consider AI utilization as part of our evaluation process,” said MacLachlan. “Instead, we focus on AI adoption and sharing best practices to celebrate innovation and operational efficiency gains across the company.”
At the same time, Andy Jassy, Amazon CEO, hasn’t been shy about his AI expectations for his employees. In a company-wide email last June, he predicted that AI-driven productivity gains would reduce the company’s corporate workforce, and urged workers to embrace AI. “Educate yourself, attend workshops and take trainings, use and experiment with AI whenever you can, participate in your team’s brainstorms to figure out how to invent for our customers more quickly and expansively, and how to get more done with scrappier teams,” he wrote.
The unspoken math
That same company-wide email prompted heavy internal pushback at Amazon last summer, with employees slamming Jassy’s leadership and speaking of the demoralizing impact of the company’s AI push, according to Business Insider. Months later, over 1,000 workers signed a petition that raised concerns about the company’s “aggressive rollout” of AI tools.
As Amazon has laid off thousands of workers, it’s shared growing revenue numbers each quarter. Though Jassy has repeatedly said that these layoffs are neither “financially-driven” nor AI-driven, for Maria, all of this adds up.
“If you say you automated away two hours of someone’s job, you need to convert that into savings on that job title,” she said, explaining the company’s logic behind cutting jobs. “That’s the unspoken math of what they’re doing.”
Jack keeps thinking about comments Jassy made during a companywide all-hands meeting last spring. According to a Business Insider report about this meeting, Jassy responded to a question about running Amazon as “the world’s largest startup”, and said they want to be “scrappy” to “do a lot more things”. He also warned that their competitors are the “most technically able, most hungry” companies, including startups “working seven days a week, 15 hours a day”.
“All of those things put together was an implicit threat that the people remaining at the company are expected to work longer and harder,” said Jack. It “really struck home to me that if [Amazon] can’t amass profits with endless growth, then it can get a little bit more by squeezing it out of the people working for it”.
