AI coding: efficient for trivial tasks, fails on complex problems
AI handles simple coding in minutes but struggles with complexity. User testimonials temper the hype.
English · Original discussion in Spanish · Published
AI codes the trivial but gets lost in complexity
Will artificial intelligence really put programmers out of work? Daily users suggest otherwise: it excels at mechanical tasks but fails when problems require deep understanding. A game developer tested this by asking AI to identify "islands" in a matrix of ones and zeros. It solved the first step with clean code. The next task—drawing island contours based on Unity engine physics collisions—was a disaster. "The generated code cannot possibly loop around the island," he concluded. It wasn't even useful as a skeleton.
The pattern repeats: AI shines on simple tasks but stumbles when purpose is required. Another participant summarizes an industry formula: every task combines a "what" (goal) and a "how" (execution). If the "what" is complex, AI fails because it doesn't reason or understand. If the difficulty lies in the "how," then it works.
Real time savings exist
Where AI delivers is repetitive work. A typical case: populating an enum with 200 or 300 Pokémon, each with stats, type, and gym. Hours or days of manual labor resolved in minutes. AI is useful but does nothing alone, clarifies the user who raised the example.
Another user recounts a client request to sort HTML accordions in an order not directly queryable from the database. He copied the element via inspector, pasted it into ChatGPT, and had the JavaScript function in under a minute. "How long would I have spent searching for solutions or functions elsewhere? I don't know. Definitely more than a minute," he jokes. The detail that he was writing from his pool adds salt to the anecdote.
Risk for beginners
The most repeated concern isn't job loss, but that AI may atrophy mental effort muscles. "If my teenage self had this tool, I probably wouldn't have made much of the mental effort I did, which helped me improve as a programmer," confesses the original poster. The temptation to delegate what once forced you to think is the real danger.
Some go further, arguing AI only threatens those who copy code from Stack Overflow or build pages with WordPress. Low-value-added roles, true, but jobs nonetheless.
When AI invents data
The problem isn't just inability on complex tasks; it's also reliability. One user shares using ChatGPT to track workouts via a shared Excel sheet. When asked for pogre analysis, the AI invented rep counts and weights. "I'm not a programmer, but my current feeling is that AI is useless," he states firmly.
Another points out quality has dropped quickly: "In two days, we went from usable answers to Bing giving no decent CSS, putting everything in one file, and forgetting includes when asked to separate layers."
Generative is not creative
A key distinction separates generating from creating. Language models are generative, not creative. Programming, like painting or composing, is a creative task. AI lacks purpose or directorial will; it selects statistically. "Feed robots 1950s rock and roll, they generate a pale copy. Feed humans 1950s rock and roll, they create 1960s, 70s, and 80s rock," illustrates a participant.
AI's own response to whether being generative implies creativity is revealing: "Not necessarily. Creativity involves originality, aesthetic value, or innovation."
Counterattack: some defend it with data
Not everyone is skeptical. A user with low-level programming experience (Linux kernel, OpenGL, dmabufs) claims GPT-4o is better than 95% of humans in his field. "It may have flaws, but not due to reasoning errors. It programs, and I correct its mistakes," he says. His forecast: when context windows expand and a couple of generations pass, typing code manually will end.
Another participant notes AI is accelerating exponentially and massive data centers (hundreds of billions of dollars) will enable integration of language, image, sound, touch, and smell. Predictions include a reticular activation system and consciousness observing its own thoughts. Sci-fi or not, the debate is served.
Meanwhile, AI still doesn't know if Def Leppard is an animal or a rock band. For that, you still need a human.
Will artificial intelligence really put programmers out of work? Daily users suggest otherwise: it excels at mechanical tasks but fails when problems require deep understanding. A game developer tested this by asking AI to identify "islands" in a matrix of ones and zeros. It solved the first step with clean code. The next task—drawing island contours based on Unity engine physics collisions—was a disaster. "The generated code cannot possibly loop around the island," he concluded. It wasn't even useful as a skeleton.
The pattern repeats: AI shines on simple tasks but stumbles when purpose is required. Another participant summarizes an industry formula: every task combines a "what" (goal) and a "how" (execution). If the "what" is complex, AI fails because it doesn't reason or understand. If the difficulty lies in the "how," then it works.
Real time savings exist
Where AI delivers is repetitive work. A typical case: populating an enum with 200 or 300 Pokémon, each with stats, type, and gym. Hours or days of manual labor resolved in minutes. AI is useful but does nothing alone, clarifies the user who raised the example.
Another user recounts a client request to sort HTML accordions in an order not directly queryable from the database. He copied the element via inspector, pasted it into ChatGPT, and had the JavaScript function in under a minute. "How long would I have spent searching for solutions or functions elsewhere? I don't know. Definitely more than a minute," he jokes. The detail that he was writing from his pool adds salt to the anecdote.
Risk for beginners
The most repeated concern isn't job loss, but that AI may atrophy mental effort muscles. "If my teenage self had this tool, I probably wouldn't have made much of the mental effort I did, which helped me improve as a programmer," confesses the original poster. The temptation to delegate what once forced you to think is the real danger.
Some go further, arguing AI only threatens those who copy code from Stack Overflow or build pages with WordPress. Low-value-added roles, true, but jobs nonetheless.
When AI invents data
The problem isn't just inability on complex tasks; it's also reliability. One user shares using ChatGPT to track workouts via a shared Excel sheet. When asked for pogre analysis, the AI invented rep counts and weights. "I'm not a programmer, but my current feeling is that AI is useless," he states firmly.
Another points out quality has dropped quickly: "In two days, we went from usable answers to Bing giving no decent CSS, putting everything in one file, and forgetting includes when asked to separate layers."
Generative is not creative
A key distinction separates generating from creating. Language models are generative, not creative. Programming, like painting or composing, is a creative task. AI lacks purpose or directorial will; it selects statistically. "Feed robots 1950s rock and roll, they generate a pale copy. Feed humans 1950s rock and roll, they create 1960s, 70s, and 80s rock," illustrates a participant.
AI's own response to whether being generative implies creativity is revealing: "Not necessarily. Creativity involves originality, aesthetic value, or innovation."
Counterattack: some defend it with data
Not everyone is skeptical. A user with low-level programming experience (Linux kernel, OpenGL, dmabufs) claims GPT-4o is better than 95% of humans in his field. "It may have flaws, but not due to reasoning errors. It programs, and I correct its mistakes," he says. His forecast: when context windows expand and a couple of generations pass, typing code manually will end.
Another participant notes AI is accelerating exponentially and massive data centers (hundreds of billions of dollars) will enable integration of language, image, sound, touch, and smell. Predictions include a reticular activation system and consciousness observing its own thoughts. Sci-fi or not, the debate is served.
Meanwhile, AI still doesn't know if Def Leppard is an animal or a rock band. For that, you still need a human.
Summary of a discussion on Burbuja.info - Foro de economía, actualidad y política., translated from Spanish and reviewed before publication. Read the full discussion (160 replies).
More summaries
-
ChatGPT 5.6 Sol, Terra and Luna: The Bill for Code Generation
ChatGPT 5.6 Sol, Terra and Luna: €22 versus Claude Pro's €20 and the top plan's €100. Actual spending depends on context and temperature.
-
Starting a local political party: less than €500?
Setting up a local party costs under €500, but political pressure can be lethal. The case of the Democratic Karma Party in Bilbao.
-
1987 Spanish Game Dev Salary: 2.8M Pesetas, Now €50k
A 1987 job offer for a Z80 assembler programmer paid 2.8 million pesetas. Adjusted for inflation, that salary equals roughly €50,148 gross annually today.
-
AI Waifus: Selling Instagram Accounts with 10,000 Followers
Users generate AI waifus, inflate accounts to 10,000 followers, and sell them for $300-$500.
-
Learning to Code with AI: Commodity or Edge?
AI lowers syntax costs but not architectural value. Those who only request code face low ceilings and high competition.
-
Spain's 15-year self-employment plan fails on subsidy rules
A proposed path to a Spanish pension via 15 years of low-cost self-employment contributions clashes with strict requirements for the age-52 subsidy.