Summary of Key Points
This news article focuses on the real dynamics of AI tools (such as programming assistants) entering the workplace of programmers: companies are generally stingy with these tools (limiting the number of tokens available and not purchasing enterprise-level subscriptions), forcing programmers to pay out of their own pockets for usage. While AI can replace routine, repetitive tasks, companies use this to raise KPIs and cut labor costs (by laying off employees or avoiding hiring new ones). AI has clear limitations; it cannot handle complex business processes or interpersonal coordination. Job roles are being redefined, with basic positions becoming obsolete, while mid-to-senior engineers and those skilled in integrating old systems becoming more valuable. At the same time, AI exacerbates the issue of ineffective competition within industries, increasing the risk of poor code quality.
1. Companies Being Stingy: AI Tools Becoming a Programmer's Expense
Companies have a dual approach to AI tools: they want the benefits without the associated costs. Either they do not purchase enterprise-level subscriptions (as in the traditional company where Shi Jian works) or they set a limited quota of tokens per person per month (as in Li Shijing's company). The reason is simple: it’s not cost-effective for businesses. The computational power required by AI is expensive, and there is no immediate return on investment; moreover, there are concerns that employees might use AI to slack off at work (large companies impose quotas to reduce token waste).
Why do programmers have to pay out of their own pockets? It’s a matter of survival. Feng Shaolin fears that using AI will make it seem like his work is being done by the tool, leading to less salary increases; Shi Jian worries about being laid off if he doesn’t use AI; Cheng Yiwei even opposes the company distributing tokens, as most companies that do so tend to cut staff.
Paying for tokens is like buying a rope to strangle oneself—by doing so, programmers can work faster without relying on others, which has become an unwritten rule of the workplace.
2. AI Improves Efficiency? It’s More About Shifting Pressure Than Reducing Burden
AI can indeed help with about 30% of routine tasks (such as checking logs and writing simple code), but the saved time is not used for rest:
- Hidden Increase in KPIs: Li Shijing’s project, which originally took one and a half months, now needs to be completed in three weeks because of AI.
- Increased Workload: Shi Jian’s workload hasn’t increased, but he has to explore more solutions or create higher-quality work, leaving him working longer hours.
- Labor Reduction: Shi Jian’s team lost two front-end developers; Feng Shaolin’s company avoids hiring new employees, so one person often covers multiple roles.
AI is not a helper but more of an accelerator—companies use it to extract more efficiency from their staff, with the savings going into the bosses’ pockets.
3. The Limits of AI: What It Can’t Do
AI has significant limitations:
1. Lack of Business Understanding: AI may pretend to understand requirements but doesn’t ask for details, leading to bugs under special circumstances (as experienced by Li Shijing).
2. Inability to Handle Complex Old Systems: Old systems contain a lot of historical data and constraints that AI cannot handle directly; manual intervention is needed to break them down into manageable parts.
3. Poor Quality in Fine-Personalized Work: User-facing products (like REDnote) require precise visual adjustments, which AI cannot perform.
4. Time-Consuming Setup: Preparing prompts and documentation for AI takes as much time as writing the code itself.
In short, AI can handle physical tasks, but the critical mental work (such as analyzing requirements, coordinating business processes, and debugging code) still relies on humans.
4. Job Role Reconfiguration: Who Will Be Eliminated? Who Will Be in Demand?
With AI, positions are not disappearing but are being redefined:
- Those Who Will Be Eliminated: Junior programmers and product managers who only know how to write basic code, as AI can take over their tasks.
- Those Who Will Be in Demand:
- Mid-to-senior engineers who review and fix AI-generated code (in high demand).
- Full-stack engineers who can handle both front-end and back-end work, with AI assisting them.
- FDE (Front-End Deployment Engineers) who specialize in integrating old systems to make AI usable (a scarce skill globally).
The entry barrier has risen; graduates now need to understand business, AI, and have critical thinking skills, essentially becoming “mini-senior engineers.” AI makes it harder for newbies to enter the field but opens more opportunities for experienced professionals.
5. AI Exacerbates Industry Inefficiencies: More Ineffective Competition
The tech industry already focuses on the number of features released as a measure of success, and AI exacerbates this problem. AI is good at quickly adding features, leading programmers to create many unnecessary ones, leaving them with less time for essential work.
The more alarming aspect is that the sheer volume of AI-generated code makes it impossible for humans to review all of it in time, forcing rushed releases and a cycle of continuous maintenance. Li Shijing fears this will lead to major issues down the line.
In Conclusion
AI is not a savior for programmers; rather, it’s a magnifying glass that highlights companies’ stinginess and the competitive pressures within industries. For programmers, they must evolve into professionals who combine AI with business knowledge or risk being phased out. Companies, on the other hand, need to address cost, management, and security issues if they want to make effective use of AI; otherwise, they’ll only solve temporary problems at the expense of long-term success.