LLMs are reshaping the self-perception of software engineers

The cognitive dissonance behind the career crisis

The technical director of a cloud service company laid off one-third of its junior engineers last week and assigned the rest a Copilot Enterprise account. The team's output did not drop and the delivery speed doubled, but he received four resignation letters-including two backbones with five years of experience. "They said they couldn't find the meaning of writing code, damn it." He took a sip of craft brew and complained: "Is it unfair for me to pay them not to write duplicate code?"

This is not a matter of technology substitution at all. When the debugger's warning sound used to be an engineer's heartbeat monitor, now there is only a silent blank terminal, and professional identity is collapsing. GitHub Copilot helped write 46% of new code in 2023, and Stack Overflow traffic plummeted 35%, but the real bomb lies in the cognitive level: Engineers suddenly discovered that the "swordsmanship" they had practiced for 20 years had been wiped out by laser guns. They were also required to use laser guns to shoot more accurately.

Efficiency Illusion Covers Career Hollow Out

Last year, an e-commerce company rebuilt its commodity system and used LLMs to generate 90% of CRUD code. The three engineers completed the original six-person-month work in two weeks, and the management announced "optimizing resource allocation" at the celebration party. One month later, the failure analysis report showed:
The average nesting depth of - automatically generated code reaches 7 levels (manual code is usually ≤4 levels)
58% of the - exception handling module directly copied and knew that the answer was not verified

• core transaction process contains three different versions of amount calculations

The engineer's roar in the conference room was very real: "In the past, I had to check the documents for three days to adjust a deadlock, but now the AI comes up with a plan in ten seconds. But when I looked at the mountain of shit it generated, I didn't dare to delete any lines. Am I an engineer or an AI tamer?"

QKPFX5 The cruel confession of the QK salary curve
Look at two sets of heart-piercing data:

1. Levels.fyi shows that the median salary of intermediate engineers in Silicon Valley will only increase by 3% in 2023 (inflation 7.1%)
2. After an IPO corporate service company used LLMs, the single-function delivery cost dropped from $12k to $4k, but customers pressed prices even harder

Capitalists calculate quickly: Have you doubled your efficiency in producing code? Then your output is only worth half price . There is a joke that spread widely in Meta: the product manager took the AI-generated PRD and asked the engineer,"Can this requirement be completed in two days?" The engineer replied,"Feed the PRD to the AI and it will be completed in half an hour." The conference room suddenly fell silent-both sides realized that they were performing the emperor's new clothes.

Steelman: What did the opponents go wrong?

Someone must jump: "Don't be pretentious! When the IDE eliminated assembly engineers, didn't the programmers live well?" These words stealthily changed three concepts:

1. Speed difference : It took twenty years from assembly to C++, and AI was compressed to twenty months
2. Mental weakness : IDE is a tool to extend arms, LLMs are a black box that replaces decision-making
3. Value difference : Debugging complex systems once brought a sense of conquest, but now it has turned into cleaning up AI's mess

What is even more insidious is the shift of responsibility. An autonomous driving team used Copilot to generate a control algorithm. After the accident, the engineer took the blame. The lawyer in court asked: "You knew that AI might make mistakes, why didn't you write by hand?" The engineer asked: "Do you believe that the handwriting error rate is higher?"

System Design Team's Choice

Wang Gong has worked in the banking system for ten years and purchased an AI code platform last year. When he revised the account verification logic generated by AI for the seventh time, he suddenly asked the team leader: "I didn't write the core algorithm of this module, I didn't set the verification rules, and even the exception handling was done by AI from Stack Overflow--if there is a mistake in fund settlement tomorrow, is it my responsibility or the AI's responsibility?"

The team leader handed him a new requirements document: "The head office requires that digital wallets be launched next month. You are responsible for training exclusive models to understand financial agreements." Gong Wang looked at the job description of "Prompt Engineer" in the document and remembered that the work card also said "Senior Software Architect".

new value coordinates are being reorganized

When writing CRUD code becomes shameful and inefficient, engineers need to re-anchor the value:

The CTO of a logistics company is very eye-catching: he feeds API design guidelines to LLMs to generate code templates, allowing people to check boundary conditions and write test cases. "Engineers now spend 60% of their time designing 'how to test AI', but the failure rate has dropped to a five-year low."

Essentially, engineers are transforming from keyboard warriors to curators in the AI era-you don't need to write faster than AI, but you need to understand why AI should generate this kind of code in a certain business scenario . Just like photographers don't have to grind silver salt particles themselves, but they must know what light to use ISO1600.

bet on a future

I bet that two new types of work will emerge within three years:

1. AI process auditor : specializes in checking compliance loopholes in automatically generated code, and the hourly salary is three times that of an ordinary engineer
2. Requirements Alchemist : Transform fuzzy business requirements into precise Tips that AI can execute

As for those who only complain that "the code generated by AI is like shit" but don't study how to train the model? Sorry, they are like carriage makers accusing cars of polluting the environment- The technological revolution never cares about the cries of old skill holders .

So back to the question at the beginning: When writing code becomes typing, what is left of engineers?
The answer is: Determine what is worth doing, define what to do right, and the courage to take responsibility for the results .