At least since the Industrial Revolution, workers have been worried that machines will replace them.
But when technology transformed auto-production, meat clogging, and even secretary jobs, the response was not usually about reducing work and reducing the number of workers. It was to “decompose” the jobs, break them down into simpler tasks, and run them over and over again with quick clips. The small shop of skilled mechanics gave way to hundreds of workers spread across the assembly line. The personal assistant gave way to a pool of typists and data entry clerks.
Workers “suggested speeding up, intensifying work, and deteriorating work,” explained labor historian Jason Lesnikov.
The same thing appears to be happening with artificial intelligence in one of the most widely adopted fields: coding.
As AI spreads into the workforce, many white-collar workers have expressed concern that it will lead to mass unemployment. However, while unemployment rates may be surged and ultimately widespread layoffs, the more immediate downside for software engineers is likely to be a change in the quality of their work. Some say it's becoming more routine, less thoughtful and crucially much faster.
Companies seem to be persuaded that AI could boost productivity, just like with the assembly lines of the old ones. A recent paper by Microsoft researchers and three universities found that programmers use AI coding assistants called Copilot. This suggested a snippet of code that can be accepted or rejected, increasing the important scale of output by over 25%.
Amazon, which has made a huge investment in generator AI, is changing the culture of coding at a rapid pace. In a recent letter to shareholders, CEO Andy Jussey wrote that generative AI is bringing great benefits to businesses that use it for “productivity and cost avoidance.” He said that if Amazon doesn't provide customers with what they need “as soon as possible,” it's essential that competitors work faster, as they gain position, and cited coding as an activity where AI “changes norms” is what they need.
Those changing norms are not always enthusiastically accepted. Three Amazon engineers said they have been increasingly pushing managers to use AI in their work for the past year. The engineer said the company has raised production targets and is not forgiving deadlines. It even encourages coders to gin up new AI productivity tools in the internal coding competition, the upcoming hackathon. One Amazon engineer said his team was about half the size of last year, but was expected to generate roughly the same amount of code using AI.
Amazon said it will conduct regular reviews to ensure that the team is staffed properly and may increase the size if necessary. Amazon spokesman Brad Glasser said:
Other tech companies are heading in the same direction. In a note to employees in April, Shopify, which helps entrepreneurs build and manage e-commerce websites, announced that “Use of AI is currently a baseline expectation,” and that the company will “add questions about AI use” to its performance review.
Google recently announced that it will soon be holding a companywide hackathon that will create AI tools that allow one category to “enhance daily productivity.” The winning team will receive $10,000. A Google spokesperson noted that over 30% of the company's code is proposed by AI and accepted by developers.
All shifts are not negative for workers. At Amazon and other companies, managers argue that AI can help employees ease their work from boring tasks and perform more interesting tasks. Jassy wrote last year that he saved “4,500 developer years” by doing his job that he appreciates using AI to upgrade old software.
Eliminating such boring research could benefit a subset of skilled programmers, said Lawrence Katz, a labor economist at Harvard University, has closely followed research on the subject.
However, for inexperienced programmers, the results of introducing AI could be similar to the transition from craftsmanship to factory work in the 19th and 20th centuries. “Things look like speeding up knowledge workers,” Dr. Katz said he explains preliminary evidence from ongoing research. “There's a sense that employers can load more.”
Bystander at my own work
Coding automation has a special resonance for Amazon engineers who are watching their blue-collar counterparts undergo similar transitions.
For years, many workers in Amazon's warehouses walked miles every day to track inventory. However, over the past decade, Amazon has become increasingly dependent on what is called a robot warehouse. There, a picker stands in one place and pulls stock from shelves delivered from lawn-like robots.
Robots generally do not replace humans. Amazon said it has hired hundreds of thousands of warehouse workers since its introduction, creating many new, skilled roles. However, the robots have increased the number of items each worker can choose from tens to hundreds per hour. Some workers also complained that the robots have over-physically taxed their jobs. Amazon says it offers regular breaks and cites positive feedback from workers about cutting-edge robots.
An Amazon engineer said the transition was in their minds. The company said that if they wanted to keep up with output targets that are technically optional but impact performance reviews, they encouraged them to rely more on AI, which said they had few options.
Expectations are rising rapidly. One engineer said it would take several weeks to build the website functionality. Now it needs to be done frequently within a few days. He said this is only possible by using AI to help automate coding, reducing meetings with colleagues and exploring alternative ideas for feedback. (The second engineer said that using AI is more modest in improving efficiency. The team is increasingly intensive with the tools.)
The new approaches coding in many companies have actually eliminated much of the time developers spend on reflecting their work. “Previously, you had a lot of slack because you were doing complicated projects. It could take a month, maybe two months. “Now, you're keeping track of everything and you can do it quickly.”
Like Microsoft, many Amazon engineers use AI assistants who suggest lines of code. However, the company recently deployed AI tools that can naturally generate most of its programs. One engineer called the tool “terrifyingly good.” The engineer said many colleagues were reluctant to use these new tools.
“It's more fun to write code than to read code,” said AI fan Simon Willison, a longtime programmer and blogger, who led to objections from other programmers. “If you're told you have to do code reviews, that's not the fun part of the job. When you're using these tools, it's mostly the job.”
This shift from writing to reading codes can make the engineer feel like he is on the sidelines at his own work. Amazon's engineers have said that managers will use AI to create one-page notes suggesting solutions to software problems, allowing them to generate rough drafts from scattered thinking.
We also used AI to test the software features we build. He said automating these features could potentially deprive junior engineers of the know-how they need to advance.
Amazon said collaboration and experimentation remains important and views AI as a tool to augment engineers' expertise rather than replacing them. It said it would clarify the promotional requirements to employees.
Harper Reid, another longtime programmer and blogger who was the chief technology officer for former President Barack Obama's reelection campaign, agreed that advances in engineering careers could become a problem in the AI world. However, he warned that he was overly valuable about the value of a deep understanding of his code.
“It's crazy to see people measuring in car factories to make sure every angle is correct,” he said. “It's not as important as it was when we were a group of 10 people pounding metal.”
And just as the surge in factories overseas has made it cheaper and easier for entrepreneurs to manufacture physical products, the rise of AI could democratize software production and reduce the cost of building new apps. “If you're a prototyper, this is a gift from heaven,” Willison said. “You can knock out something that shows you an idea.”
A scary speed-up
Amidst their frustration, many Amazon engineers joined Climate Justice's group called Amazon Employees. (Amazon said it is working to reduce carbon footprint from data centers. The Climate Justice Group is pushing it to provide more information on how to do so.)
Group organizers regularly communicate with hundreds of Amazon employees, saying workers are increasingly discussing the stress of using AI for work, along with the effects of technology on the climate.
Eliza Pan, a former Amazon employee who spokesperson for the group, said the complaints were focused on “what their careers look like.” “And not just their careers, but the quality of their work.”
There's no rush to form a coder union on Amazon, but such a move is not unheard of. When General Motors workers took a strike in 1936 to demand recognition of the union, the coalition auto workers, it was a horrifying speedup that spurred them.
A typical worker felt that “he was not free because he was probably in his previous job, so he was probably in his previous job to set the pace of his work.”
