Xinzhiyuan Report
Edited by: Ying Zhi
【Xinzhiyuan Introduction】As AI tools accelerate code generation, programmers' creativity faces unprecedented challenges. Is AI a magic wand that frees our hands, or a cage that stifles thought? Discover the true feelings of Amazon engineers!
Will the future of programming be a perfect balance of inspiration and efficiency, or an assembly line code factory?
Recently, software engineers at Amazon have found that their daily work is undergoing subtle yet profound changes.
AI tools are widely introduced, from code generation to debugging optimization, and programmers are required to deliver results at a faster pace. Sounds cool, right?
AI writes code, efficiency doubles! But the reality is not that simple.
Some programmers complain that the intervention of AI makes them feel that time is compressed, thinking space is squeezed, and the work pace is increasingly like a warehouse worker sorting packages: fast, mechanical, repetitive.
Behind this change is Amazon's endless pursuit of efficiency. As a global e-commerce giant, Amazon's warehouses are known for their highly efficient automated processes.
Now, this "assembly line thinking" seems to be permeating the field of software development.
Programmers are required to use AI tools, such as code completion assistants and automated testing frameworks, to shorten development cycles.
What's the result? They find themselves no longer creators "from scratch," but more like workers assembling code on an assembly line.
In the past, a complex project might take a month, or even two months. Now, the entire process is monitored and can be completed quickly.
Since the Industrial Revolution, the anxiety of machines replacing humans has never dissipated.
Historical experience shows that the more common impact of technological change is not layoffs, but job degradation—breaking down complex tasks into mechanically repetitive simple operations.
Previously, car factory technicians were led by master craftsmen. Later, with the introduction of assembly lines, everyone became a tool for tightening screws: repeating the same action hundreds of times a day.
Machines did not directly make people unemployed, but they broke down all tasks into brainless, repetitive labor.
Creativity or Productivity?
Now this trend has reached programmers.
Everyone was originally worried that AI would take their jobs, but it turns out what's more annoying is that the work hasn't decreased; instead, it has become faster and rougher.
Programming should be a carnival of intellect and creativity.
Good programmers not only write runnable code but also design elegant architectures, anticipate potential bugs, and leave room for code expansion.
However, with the "assistance" of AI, opportunities for such deep thinking seem to be diminishing. Engineers worry about losing important skills and promotion opportunities as a result.
At Amazon, management's advocacy for AI is almost fanatical.
They believe that AI can not only improve efficiency but also stabilize code quality.
Amazon CEO Andy Jassy stated that AI has saved them thousands of years of development time.
However, programmers have different feelings.
Some complain: "AI-generated code is like fast food; it fills you up, but it lacks flavor."
Even more, they feel their work is being degraded, transforming from highly creative intellectual labor into mechanical code portering.
Amazon's CEO clearly stated in his shareholder letter: Using AI can make programmers' efficiency soar, and those who are slow will be eaten by competitors.
He believes that "speed" is key to maintaining a competitive advantage, and generative AI can save a lot of costs.
The engineers below are suffering: the team size has been cut in half, the code volume requirements remain unchanged, and they are struggling to cope relying entirely on AI-generated code blocks.
Some Amazon engineers admit that they are now more like piecing together AI-generated code snippets rather than designing solutions from scratch.
The company recently launched AI tools that can generate most programs themselves. One engineer called these tools "terrifyingly good."
Some said many colleagues are reluctant to use these new tools because they require a lot of repeated checking, and engineers want more control.
One colleague said that previously, a new feature would take two weeks to develop, but now it has to be delivered in three days, with frantic copy-pasting every day, leaving no time to discuss solutions with colleagues.
AI generates code quickly, but it always feels like it lacks a "soul."
What is more worrying is that this high-intensity pace may be stifling programmers' creativity.
In the past, programmers had time to delve into a complex problem, or even spend days optimizing an algorithm.
Today, the rapid output of AI tools has raised management's expectations for delivery times.
Programmers are forced to complete more tasks in less time, with thinking time compressed to a minimum.
Someone joked: We are not writing code now, but racing against AI!
Code that previously required weeks to develop now has to be delivered within days. Programmers must rely on AI to keep up with project progress, otherwise, their performance will be affected.
Of course, the introduction of AI is not entirely bad.
Code completion, automatic debugging, and even generating entire functions—AI indeed makes some repetitive tasks more efficient. Especially for for beginners or projects that require rapid prototyping, AI tools are a lifesaver.
Programmers Become "Reviewers"
Amazon's story is just a microcosm of the tech industry.
With the popularization of AI, more and more companies are beginning to rely on these tools to accelerate development processes.
Shopify directly writes "ability to use AI" into performance reviews, and Google is even tougher, holding an AI productivity tool development competition, with a $10,000 prize for the winner.
Data shows that 30% of Google's code is now automatically generated by AI, and programmers have transformed from creators into "reviewers."
But this also raises a profound question: As AI takes over more and more programming tasks, what will the future of programmers look like?
Will they become more efficient creators, or be trapped in an assembly line-like cycle?
Amazon management stated that AI helps handle tedious low-level code, allowing programmers to focus on high-level work such as architecture optimization and algorithm upgrades.
By using AI to complete the thankless task of upgrading old software, the company saved human resources equivalent to 4,500 developers per year.
Amazon stated that AI is meant to enhance engineers' professional capabilities, not replace them, and collaboration remains important.
For senior programmers, not having to waste time writing "hello world" indeed improves efficiency.
Just as the proliferation of overseas factories enabled entrepreneurs to cheaply and easily manufacture physical products, the rise of AI may democratize software development, reducing the cost of developing new applications.
The result of introducing AI may be similar to the shift from manual labor to factory labor in the 19th and 20th centuries.
But newcomers are suffering—they used to practice by writing test code and debugging interfaces, but now AI handles everything. Many junior engineers complain: "There's no chance to debug, how can we learn real skills?"
AI can be a tool to free up hands, or a shackle that suppresses thought.
Programmers look at the robots in the warehouse, as if they see their future selves.
Previously, Amazon warehouse managers walked tens of kilometers daily to find goods. Now they stand still waiting for robots to bring shelves, and although they don't walk, the hourly sorting volume has increased from 30 items to 300, making them too tired to straighten their backs.
AI helps write code faster, but reviewing hundreds of lines of automatically generated code every day makes their eyes glaze over, completely like assembly line quality inspectors.
Worrying Acceleration of Pace
An internal Amazon group, originally protesting the company's carbon emissions, has recently become a place for programmers to vent their frustrations.
Eliza Pan, a spokesperson for the organization and a former Amazon employee, said these complaints primarily revolve around what their careers will look like, not just their careers, but also the quality of their work.
The shift from writing code to reading code makes engineers feel like bystanders in their own work.
Hundreds of people chat in groups daily: "Will using AI to write code mean I'll have nothing to put on my resume later? The boss only looks at code volume, not whether the logic is optimized; what technical content is left?"
Someone mentioned the 1936 General Motors strike; back then, workers were also pushed too hard by the assembly line. Now, programmers seem to be approaching that point.
Of course, there are optimists too.
Someone gave an analogy: "In the past, cars were made by blacksmiths hammering iron sheets; now they are stamped by machines. Can we say car manufacturing technology has regressed? AI merely standardizes basic work; truly skilled engineers should focus on innovative design."
For startups, AI is simply a lifesaver.
Previously, it would take 10 programmers to set up an APP framework; now, with AI tools, two people can create a prototype in a week.
In this AI-driven era, the role of programmers is being redefined.
They need not only smarter tools but also more time to think, to create, and to give life to code.
Someone joked: What should we ask in interviews in the future?
Not "Can you write algorithms?" but "Can you quickly review AI-written code?"
As for whether this is good or bad—just like when assembly lines first appeared, some grumbled, some quietly adapted, and eventually everyone had to follow the times.
Behind technological progress, there is always a test of humanity and creativity.
When AI Enters Programmers' Daily Lives
Researchers from institutions such as Princeton University and MIT conducted large-scale field experiments at Microsoft, Accenture, and an anonymous company, aiming to explore the impact of generative AI on software developers' productivity through real work scenarios.
This study focused on GitHub Copilot, an AI coding assistant developed by GitHub in collaboration with OpenAI, which can generate code completion suggestions based on context and has been used by over 1.3 million users and 50,000 enterprises.
The experiment covered nearly 5,000 software developers, including 1,746 from Microsoft, 320 from Accenture, and 3,054 from the anonymous company.
These personnel covered various positions from junior to senior, with tasks including code writing, testing, and project management.
Researchers tracked three core metrics through GitHub's version control data:
Task completion (number of pull requests): Measures independent units of work completed by developers, such as new features or bug fixes.
Code activity (number of commits): Records the frequency of code modifications, reflecting the iteration efficiency in the development process.
Compilation efficiency (number of builds): Evaluates the number of successful code compilations, indirectly reflecting code quality and the smoothness of the development process.
Overall Efficiency Increased by 26%
Researchers found that developers using Copilot increased their average weekly task completion by 26.08%, code commit frequency by 13.55%, and compilation frequency significantly by 38.38%.
This indicates that AI assistants not only accelerated task completion but also spurred more frequent code iterations and testing.
The experiment revealed an interesting phenomenon: less experienced developers had higher acceptance of Copilot and showed more significant productivity gains.
Junior developers' output increased by 21%-40%, while senior personnel only saw an increase of 7%-16%.
Why do novices benefit more from AI?
Novices are more willing to accept Copilot's code suggestions, viewing AI as an intelligent assistant to fill knowledge gaps.
Novices often learn through a "try-compile-adjust" loop, and Copilot's real-time suggestions reduce ineffective attempts.
The surge in compilation times (+38.38%) reflects their more frequent validation of AI-generated code, and the build success rate did not significantly decline, indicating that the overall quality of AI suggestions is controllable.
Despite Copilot being available without additional investment, 30%-40% of developers never tried it.
Experienced developers rely more on the sense of control from manual programming and believe AI might disrupt code style consistency.
Some developers worry about security vulnerabilities or copyright risks in AI-generated code, especially when dealing with sensitive projects.
Enterprises should specifically promote the adoption of AI tools, for example, by providing training for novices, encouraging teams to use Copilot for repetitive tasks (like code template generation), and unleashing the innovation of highly experienced employees.
Novices can use AI to accelerate basic programming, freeing up time to learn advanced skills like architecture design; senior developers should focus on areas where AI is hard to replace, such as complex system optimization and requirements analysis.
As AI technology iterates, how to balance tools and human creativity will become a proposition that all knowledge workers need to consider.
References:
https://www.nytimes.com/2025/05/25/business/amazon-ai-coders.html
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4945566