Summary of Key Points
Artificial Intelligence (AI) has significantly accelerated the speed of software development, but it has also magnified issues such as technical debt and system complexity. The value of engineers has shifted from simply writing code to their ability to make informed decisions. The "middle layer" of software engineering, which relies on execution skills, is facing pressure. In the future, the bottleneck in AI programming will shift from generating code to understanding it, a trend that could extend to all knowledge-based work. As a result, judgment will become increasingly valuable, while the value of execution skills will be diluted by AI.
Detailed Analysis
1. AI speeds up code generation, but technical debt grows like an "invisible credit card bill"
In the past, slow code writing was a natural constraint; adding new features required careful discussion of architecture and database design to avoid making random changes. Now, AI (such as agents) can quickly generate large amounts of code—20,000 lines of proposed changes (PRs) in just one weekend. On the surface, the functionality seems usable, but the underlying problems are more hidden, much like buying a luxury car with a credit card, only to realize you've accumulated significant debt later on.
For example, someone might use AI to create multiple database tables for convenience, but later, when changes are needed, the cost of data migration and consistency issues could be ten times higher than the initial code generation. Worse still, the person who wrote the code may not understand why AI made those choices, and any bugs would have to be resolved by asking the AI again, with no one able to verify the accuracy of the AI's decisions. Technical debt is no longer a "manageable loan"; it has become an "invisible liability whose amount is unknown."
2. The value of engineers lies in their ability to decide what code to write and what not to write
Engineers in Silicon Valley are highly paid not because they can simply convert requirements into code (outsourcing is often cheaper), but because they can make informed decisions, such as whether to include Kafka in a system or whether the database design will affect future scalability. AI can help with code writing, but it cannot make these strategic judgments. It will generate code based on given requirements without understanding the long-term consequences.
For instance, AI might suggest adding a new service, but a skilled engineer would consider whether the service is truly necessary and whether it will increase system complexity. In the future, engineers who can make such assessments will be more valuable as they can leverage AI to enhance their own contributions, while those who merely rely on AI for coding will become less valuable.
3. The "middle layer" of software engineering is at risk
In the past, middle-level engineers (with some experience and basic coding skills but lacking decision-making abilities) could earn decent salaries by focusing on execution tasks. Now, AI can handle most of the execution work, leaving this group in a difficult position:
- Top-tier engineers: They use AI to complete tasks quickly and focus on making critical decisions, thus increasing their value.
- Less capable engineers: AI may cause them to make more mistakes (e.g., generating 20,000 lines of poor-quality code per day), leading to higher costs for the company.
- Middle-level engineers: If they cannot improve their decision-making skills, they will struggle to compete with both top-tier engineers and the efficiency of AI.
4. The next challenge in AI programming is understanding the generated code
AI can generate code at speeds beyond human comprehension; agents can modify dozens of files simultaneously, but humans need hours to understand the dependencies between these changes. Systems are becoming increasingly complex, and no one can clearly explain how they function. In the future, the focus of AI programming will be on helping humans understand the code—tracing the decision-making process behind it, explaining its logic, and controlling system complexity. Tools that can address these issues will be more valuable than fast coding agents.
5. Judgment is crucial for all white-collar workers, not just programmers
The implications of this are not limited to programmers but apply to all knowledge-based workers:
- Doctors: AI can provide diagnostic suggestions, but doctors must determine their accuracy.
- Designers: AI can create designs, but designers need to choose the most suitable solutions.
- Marketers: AI can generate copywriting, but marketers must decide which messages will resonate with customers.
AI enhances execution skills, but judgment remains irreplaceable. In the future, those who possess strong judgment will be the most valuable, while those who rely on execution will face significant pressure.
In conclusion, AI makes it cheaper to perform tasks, but making the right decisions becomes more expensive. Regardless of the profession, the ability to determine what to do and what not to do will be the key to future success.