Do Not Let AI Write Everything: A Haskell Developer on Keeping the Joy of Code
Haskell（HASH-kell）
A programming language for writing software.
LLM（L-L-M）
An AI system that makes text or code.
Hacker News（HACK-er news）
A website where people share technology stories.
What happened
On September 18, 2026, a user named turion published an essay called How to keep enjoying programming in a world of LLMs on the Haskell Community forum. The original post is a personal argument, not a research study. It asks how programmers can use AI without losing the pleasure and skill of programming.
The author says large language models, or LLMs, can produce code quickly. But large amounts of generated code may be hard for a person to understand. That can make bugs harder to find. It can also reduce practice. The author therefore suggests a clear split: humans write important code, while AI handles planning, note-taking, research records, routine cleanup, and low-risk changes that are easy to check.
Background
The essay does not call for total AI abstinence. It challenges the idea that the only choices are full automation or no AI. The author wants AI to work around the central craft of programming. An AI tool can turn a long discussion into tasks. It can organize notes. It can list places that may need edits. The human should still understand the subject and make crucial decisions.
The author also advises researching in parallel. Do not accept an AI research result as a fact without checking it. The point of the tool is to reduce tedious searching, not to replace human understanding.
Why it matters
This argument changes the question from whether AI is allowed to which work should remain human. For the author, programming is not only a way to ship software. It is also practice, communication, and a source of enjoyment. Keeping the human involved can make the code easier to explain and maintain. It can also expose a bad plan before an automated coder spends a long time following it.
The essay recommends an automated review cycle. AI-made code and plans should receive another check before a person accepts them. It also recommends keeping useful files and a task list outside the AI session. That gives the programmer something to do when a service reaches a usage limit or becomes unavailable.
What the source confirms
The source clearly recommends keeping some hands-on coding in the workflow. It suggests using coding tools for cleanup, small routine tasks, and low-risk refactoring. It warns against handing over complicated design work without understanding the result. It also asks people to keep human-to-human communication human. A fully generated pull request may contain correct details, but it may not show that the sender understands the change.
The essay appeared on Hacker News, a technology discussion site, and received 66 points and 103 comments in the supplied listing. The Hacker News thread shows community attention. It does not prove that the essay's advice works.
The author reports being more productive, perhaps twice as fast. That is a personal estimate, not an independent measurement.
What remains unknown
The source does not compare this workflow with full automation or manual coding. It gives no error rates, time records, or team-wide results. It also does not show whether the same balance works across languages, jobs, or companies. The effect of automated review is not measured either. One developer's experience can suggest a question, but it cannot settle it.
What to watch next
The useful test is not how many lines an AI produces. It is whether people can explain, test, and maintain the result. Teams may need clearer rules about who makes important design choices. They may also need plans for work that continues when an AI service is unavailable. The essay's main lesson is a practical one: use AI for the work around programming, while keeping enough programming in human hands to preserve understanding and responsibility.
How to use AI without losing the fun of coding
📰 Full story: Do Not Let AI Write Everything: A Haskell Developer on Keeping the Joy of Code
A Haskell developer suggests using AI as help, not as the whole programmer.
Haskell（HASH-kell）
A language for writing computer programs.
LLM（L-L-M）
An AI that makes text or code.
Hacker News（HACK-er news）
A site where people share technology stories.
💡 The gist
- Keep writing important code yourself.
- Let AI help with plans and dull tasks.
- Check AI work before trusting it.
On September 18, 2026, a writer named turion shared an essay. It appeared on the Haskell Community forum. Haskell is a programming language. The writer enjoys using it.
The writer worries about giving all coding work to AI. A large language model, or LLM, can make text and code. It can make code quickly. But generated files may be hard to understand. They may not match the person’s own way of thinking. Then finding a bug can take longer.
Writing code is also practice. The writer says that stopping practice can weaken a person’s coding skill. So the writer suggests a mixed approach. People should write the important parts. AI can make plans, sort notes, record research, and do small repeated tasks.
The writer says people should research beside the AI. They should check important facts themselves. They should not let AI make crucial choices without human understanding.
The essay also suggests an automatic review. A second checker can look at AI-made code or plans. This check should happen before people trust the work. The writer also recommends keeping a task list outside the AI tool. That list helps when the tool reaches a usage limit.
The essay reached Hacker News, a technology discussion site. It received 66 points and 103 comments. Those numbers show attention from that community. They do not show that the advice is correct.
The writer says this method made work faster. The estimate is perhaps twice as fast. That is one person’s experience, not a scientific test. We still do not know if it works equally well for other people or teams. The key question is simple: can humans still explain and fix the code? If yes, AI may help without taking away all the learning and fun.
💬 How Can LLMs Help People Keep Enjoying Programming?
Some commenters feel LLMs remove boring work. Others worry that they create code people cannot understand. The key question is what the human still checks and decides.
- Users said they use LLMs to inspect data and logs, make graphs, write tests, and explore large codebases. They reported finding small problems and gaining more time for design, but these are personal experiences, not guarantees.
- Some people use an LLM only as a powerful search helper and keep it away from their code. Others let it handle chores they dislike and write the enjoyable parts themselves.
- Supporters like spending more time on architecture, interfaces, and product decisions instead of boilerplate and documentation searches.
- Critics say an LLM is not like a compiler: a compiler follows defined rules, while an LLM can give different answers and still needs careful review.
- Users reported that large, poorly understood AI-generated changes can solve the wrong problem and add workarounds that make later review harder.
initial digest at 194 comments (revision 2). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
Let AI help, but keep writing too
📰 Full story: Do Not Let AI Write Everything: A Haskell Developer on Keeping the Joy of Code
A person who writes computer instructions wants AI to help without taking all the fun.
Haskell（HASH-kell）
A language for giving instructions to computers.
AI（A-I）
A computer system that can help with tasks.
Hacker News（HACK-er news）
A place for sharing technology stories.
Haskell is a computer language. A person named turion likes writing it.
The person also wants help from AI. AI can make plans. It can tidy notes. It can do boring, small jobs.
People should write the important instructions themselves. Then they know what the computer does. Writing also helps them keep learning.
AI-made instructions need a careful look. A person should check them before using them. People should keep a list of jobs too. The list helps if the AI stops working.
This story appeared on Hacker News, a place where technology readers share stories. It got 66 points and 103 comments.
Those numbers show attention. They do not show that the idea is true.
We do not know if this plan works for everyone. The main idea is easy: let AI help with small jobs, but keep doing important work yourself.
💬 LLMs and Programming
An LLM can help write computer instructions, but a person still needs to check them.
- Users say they may ask it to do boring jobs, like making tests or reading many files.
- Some people write the fun parts themselves and let the LLM do chores.
- Users warn that an LLM can make mistakes, so a person must read and test its work.
- Different programmers enjoy different things: using help, or writing every part by hand.
initial digest at 194 comments (revision 2). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.
💬 How to Keep Programming Enjoyable in the Age of LLMs
Hacker News commenters split between seeing LLMs as a way to remove drudgery and seeing them as a source of code nobody fully understands. The recurring answer was that enjoyment depends on what is delegated and who retains responsibility.
initial digest at 194 comments (revision 2). We fetched 100 comments and sampled 100 across the thread. These are HN users’ reports, not independently verified facts.