Ask HN: Does it still make sense to write code by hand?

It's like any other tool. Use it when you need it.

We don't insist that HVAC installers use hand-powered drills to "feel closer to the work," but we also don't insist that they use battery-powered drills and impact drivers on every single task.

We don't insist that a chef use a Vitamix blender to chop a single onion when a couple seconds with a sharp knife will do, but we certainly wouldn't only use a sharp knife to create a salsa.

We don't insist that a carpenter use only hand-saws to build a house, but we also don't insist that they use a chainsaw to make a chair.

We don't insist that an accountant write out taxes by hand, but we also don't insist that they use a calculator to determine how old you are from your birthdate.

You use the power tools when appropriate, and you use more delicate, surgical tools when appropriate. It's not any more difficult than that.

The problem with this analogy is an LLM isn't very much like a "tool" at all.

- In cases of practical relevance, you pay a recurring fee to keep using it - not very tool-like.

- Rather than doing a predictable action in response to user inputs, you give it imprecise instructions and it does something unpredictable but vaguely correct in response, which then must be thoroughly inspected.

- In cases of practical relevance, its actual operation (not just design) is entirely handled by a third party, not by the user.

These things describe a subcontractor, not a tool. We don't insist that a chef at a restaurant use this tool or that tool, but if the restaurant was actually just outsourcing all its cooking to some other company...yeah, that would absolutely factor into my decision to go there or not.

Sure. But this is a little pedantic, no?

Let's go with your analogy: you would hire contractors to build a house, but not to put an IKEA bookshelf together. You would hire a professional chef to cook a five course meal, but not to heat up pop tarts. You would hire a surgeon to operate on your brain, but not to prescribe advil.

The point is that it's one of many things (tools, contractors, whatever) in your arsenal, and you use whichever one you need. You don't automatically reach for the surgeon or the chef or the contractor.

But the thing is, in this analogy, we are the chef, surgeon, or builder. We're the ones other people look for when they can't do the thing by themselves (in this case, writing software).

Neither my analogy or yours is exact, but it's really worth thinking about.

Right. But OP's question isn't "do you need a chef?" the question OP posed is more akin to "why should chefs use knives when they have blenders?" or "why should I build an IKEA bookshelf myself when I can hire a contractor to build a house?"

Because they're two different tools and situations.

You may be right but it seems like a different type of tool when it does the thinking for you. I feel like your analogy is closer to a good autocomplete rather than an agentic LLM loop. Agents feel like a genuinely different technology.

Maybe it’s more like an HVAC installer has a tool which tells them where to position the furnace. After many years, will they forget why the furnace should go there? Or maybe it’s like the HVAC installer has a tool which auto attaches the pipes. Hopefully it attached cold to cold and hot to hot, but the HVAC installer hasn’t had to do this in a while so they kind of forgot how to do it manually, and so will just trust the tool.

Imagine a carpenter needed to be proficient in hand-saws to use their chainsaw well, and using their chainsaw required knowing details about the wood they were working with, in only the way using a hand-saw gave.

Makes it a littttttle more complicated.

Conversely, imagine only requiring carpenters to know how to use a chainsaw.

I really like your analogy as-is, but LLMs aren't reliable power tools. Asking for external help requires good communication and can be overkill for small tasks.

Hiring a team of subcontractors carries risks unless you babysit the whole project.

I mean, it's the same idea, no? You hire contractors to build your house, but not to put together an IKEA bookshelf.

...and it's fine to not use the loud, vibrating power saw and instead take some extra time to use a handsaw, and thus get more practiced at it.

It's fine to hire movers, but sometimes they end up damaging stuff. My solution is have a lot less stuff, and move it myself.

I increasingly have more respect for the Amish community one of my relatives lives near, at least in terms of their considered, intentional limits set on resource use. I don't agree with all their choices, and some seem arbitrary, but I respect their autonomy to choose.

I choose to move around the world primarily under my own power or augmented by a battery (supplies 2/3 of the Wattage I need to make my cargo bicycle move); harm reduction in the face of fossil-fuel addiction. I choose not to use generative AI because I'd rather human do more work for themselves instead of selling short future humans with profligate resource consumption.

It's fine to strive for things, but the most important thing in my life is my relationships with other humans. If I'm a person who can't walk, and am in physical therapy towards being able to walk, it's fine to keep working at it, and it's fine to never be able to walk; what matters is how I get on with my mates, and spending massive amounts of electricity on data centers is arguably harmful to the health of interpersonal relationships.

This is a great frame. Can I ask (and I mean this with good faith, because I've had this problem for the longest time now) how do you deal with employers or corporations who demand the absolute maximum output every second of every day, forcing you to use the proverbial loud, vibrating power saw to the point that you feel terrified to even look at the handsaw?

The question is what your tolerance to this kind of thing is. If the job is fundamentally changing underneath you, then you need to ask yourself if this is still a job you want to keep. If the answer is no, then you find another one. If you think you can still do it the way your employer is demanding, then stick around.

The sad truth is that the job is changing, and that means we need to change, too. That _doesn't_ mean we need the power saw 100% of the time, but it does mean we need to be very very proficient at it or we'll be left behind.

Yes, exactly.

> I believe that writing something yourself helps you learn it better, but at the same time, if you did all the thinking and now just prompting for a fancy auto-complete, we could argue that the hardest part (the design) was already done, and you skip the writing for the sake of writing. On the other hand, I am afraid that if I will stop writing code in environments where I can still do it, I will lose the ability to think about code and understand code.

Merely completing the "design" is by far not "doing all the thinking," and this is really the core of the problem. When people code with LLMs, the LLMs are making nontrivial decisions and thus are replacing part of the programmer's intelligence. That's why calling LLMs "fancy autocomplete" is so misleading, because autocompletion isn't supposed to make decisions, just to facilitate typing out the code you already have in your head.

So yeah, I agree with the last sentence about losing the ability to understand code, and that's a sufficient reason to avoid AI assistance whenever possible. Since I prioritize education and learning above almost everything else, it's not a close call. Where the line gets blurry for me is when people say "I would never have learned the tech stack needed to make this tool myself, but I vibe-coded it and now it's mine." It's really difficult for me to argue with that, to tell someone simply to live without something that's potentially improving their day-to-day life. To be clear, there are many good reasons not to do that thing, but it's also not my place to judge people for doing it, since I don't live my own life perfectly either.

> but rather prompting very small tasks after you have made the initial design, something along the lines of "Create a database migration that adds column [x], not null", "add a field to the this form that will accept string value, validate it on the backend against this format, and save it to column [x]". So basically, writing code in plain English.

I find such tasks easier to just do in the programming language directly than describe them in English.

> you do it on your own terms, rather than on the terms of the employer

You've setup a condition where essentially every answer is true. I know someone who spins wool and dyes it and makes socks. Does it still make sense? Only once you add in the fact she enjoys.

If you're real question is what can be done better by hand in July 2026?

afaik optimization problems are still getting superior results from humans. Things like if you do this algo you can loop the array once instead of 3x. If you align this struct like this it's faster on a certain arch + os. Or if we do these 2 steps in microservice A we can only touch the memory and network once instead of twice etc.

a 2x speed up is still meaningful at scale. (I'd bet there are 100s of software companies out there that would pay 7 to 8 figures if you could 1/2 their AWS bill)

> If you're real question is what can be done better by hand in July 2026?

The answer to that question is "everything". Humans are still noticeably better than LLMs at writing code, and just as fast once you factor in the time it takes to review and correct the LLM.

> just as fast once you factor in the time it takes to review and correct the LLM

Do you have a source for that claim?

Not realy, using AI helper is saving time and no one will blame you for this. Today AI is our new reality and it will be self-developed in the future and become absolutely normal. Just imagine is such websites like https://maxijournal.com would be hand-made, how much time it would take for an owner to grow the website and how he could compete with others who develop such websites with AI? I can compare it with a stick-shift. Since there is an automatic transmission what good is to use a manual one, and who would blame?

I am a solo developer. I write most of my code by hand. I use LLMs to help me do research and to help find stuff in documentation. I also use it to do come reviews since I don't have anyone else to review the code. I do get a lot of value out of the code review. It often finds little things I missed or will tell me if my code sucks.

Really for what I am doing writing the code is never the bottleneck. I spend little of my time actually writing code. Where I spend most of my time on the design and requirements.

I do all of my code by hand. I don't even use an LSP. It makes sense as long as it makes sense to you.

You don't need to change the way you do things because of some perceived thing you're missing out on. That is what the AI companies want you to feel and think because it benefits them.

This applies to more than AI, but programming language choices and more. Heck, it even applies to social issues and politics :)

Make your own choices and walk your own path.

I am already being forced to use Claude for all coding in my job. All personal projects/code I still write by hand. This has always been the fun part of software development for me.

Which is great and why I started coding too. But for my day job I can’t have an employee deliver 1/4 of the code of someone using AI and ask for a full paycheck.

This, the FOMO is real, but giving in to it based on others perceptions is naive

Would you ever want to be as helpless at a skill you care about as the typical person is about math, since calculators were invented?

> prompting very small tasks after you have made the initial design, something along the lines of "Create a database migration that adds column [x], not null", "add a field to the this form that will accept string value, validate it on the backend against this format, and save it to column [x]". So basically, writing code in plain English.

Are you presenting this as the option that allows you to have better understanding (smaller tasks, closer to the changes), or the option with which you don't have understanding (because you're still getting assistance, not solving every problem on your own and not typing the code out yourself)? I genuinely can't tell from your post, people can and do take both of these positions.

> I will lose the ability to think about code and understand code

If you lose it (whatever "it" is), it's because you don't use it. If you don't use it but you're still getting your job done, is that really a loss?

Yes, there is still a need to learn to write code by hand.

Maybe not to the detailed nitty-gritty level in a computer language that was needed pre-LLM era but nevertheless still a must.

In this context, it is good to revisit the practices from early days of programming where "System Analysts" and "System Designers" provided detailed specifications to "Programmers" who translated it into computer code. The key point to understand here is that the specifications were detailed i.e they were code just in a non-machine executable form.

So detailed Specifications followed by Verification is mandatory. Understanding the code implementation in-between has lesser need but still important.

See also my previous comment chains from here - https://news.ycombinator.com/item?id=48800484

In the context of coding on your own terms, I think it makes sense to still code to help learn and reenforce understanding. There are a lot of times I read something or see code, I think I get it, but it’s not until implementing it, and making changes to that implementation, that I actually understand it.

There is also a question of enjoyment. While you enjoy it, and while you’re learning, write it yourself. If you’ve already done something many times, understand it well, and it’s become tedious in a bad way, maybe that’s where some strategic use of AI comes in. Of course, I tend to like a bit of tedium from time to time.

Its rare for me now. Most by-hand stuff I do is setting up initial repositories and folders for the base structure I want, copying logos, assets, images, content, and then I tell AI to finish the rest for example.

Configuring secrets for CICD, .env files, hosting configurations/backups.

Writing instructions files, specs, copy/pasting context, requirements, examples.

I see the argument here constantly "But ur skills!". OK. Well back to work actually shipping software using this new tool that is never going away. My client doesn't really care about my skills, just my speed and they have money for me.

I write all production code by hand, it's the only way I'm sure I know what the code does. Moreover, when the problem space or code base is more complex, it is necessary to keep in touch with reality and test your understanding by writing the code.

I see colleagues using LLMs more and more, and the number of bugs found by QA is increasing dramatically, because the LLM was not thorough enough and the developer (now reviewer I guess) missed the edge cases too because the code, tests and comments looked convincing.

I still think I haven’t gotten much more productive than a year ago:

I wrote the barebones, scaffolding, and where I cared about the details. Agents stamped out the rest from those patterns.

Everyone speaks about this stuff in anecdotes. Or heroic attempts to have agent only projects.

But very little actual data on real gains in actual quality as experienced by users? Harder to measure IMO. And may even be subjective to each developer.

Yeah, still write by hand.

Even though the translation from English algorithm to code is the "boring" part that an LLM can do, the process of doing that works as a kind of REM sleep where I'm coming up with ways to make the program better

And for a lot of simple tasks, there's plenty of stuff where it's quicker to type a command than an English description of the same thing.

It makes sense to write code by hand.

We are in a hype bubble and it can be hard to go against the grain. The bubble will calm down, as it has every other time: crypto, low code, outsourcing, SPAs, Java, AI that last time around (search for Lighthill report).

I agree with your concern about skill stagnation or erosion. Years ago we used to make fun of "architectural astronauts": developers who stopped implementing and instead used English to design systems for others to implement.

I'm guessing you have had the experience writing code where you throw away the first, second and maybe third implementations? While you "work out" what the real problem is? Our human brains need that niggling at the problem before we can make a good design. Don't short circuit it.

I think you could make a high level brainstorm session / plan on paper of things to prompt and that this was valuable even before AI

and occasionally manually writing out the code could be useful as a mental exercise to keep coding skills sharp?

I have a related question: In a contest where all the code is generated by LLMs and not hand-coded anymore, does it still make sense to use high-level languages and other abstractions that were created for humans?

I believe so, with Rust being exhibit 1. Rust is known for finding rather a lot of bugs while compiling instead of during debugging. The quote you'll often see is: "If it compiles, it usually runs."

The Rust compiler enforces many code style conditions that you most probably want in your finished program... Type checking, only one thread writing to memory at a time to avoid race conditions, etc.

It is possible to imagine an AI version of Rust that takes this even further in the future.

Maybe not the ones created for humans, but high level languages, probably. More abstract languages ("high level" basically means more abstraction away from the underlying machine, with "low level" being closer to the machine) allow for compression relative to unabstracted languages.

LLMs are known to do worse the more code they have to go through. Therefore, using a high level language where it doesn't have to work with as much code is going to be far more effective.

Turns out I think I always liked the end product more than getting there. I would kind of hate having to only write code by hand for a paycheck again.

I still write code by hand because sometimes it's faster and cheaper.

If it’s a preference question then certainly yes, but how to push back on x productivity boost argument when a company enforce this as policy?

Yes a soft policy but it’s a smart move to encourage the rat race.

lot of times I feel the code that is generated has better readability and better documentation then if I had coded. so if something is doing better job then I will use that, but reviewing the code that is generated I feel is still my job, sometimes code generated is not correct, so I need keep checking.

Outside of work (where they force me to use the clanker), why on God's green earth wouldn't I write code by hand? That is the fun part; I'm not going to voluntarily give up the fun part of my hobby project. If you are a person who is wired to care more about getting to the end than the process (as some here have said they are), then fine, use an LLM if you want. But there's no "makes sense to" involved here - if you like writing code then write code, if you don't then don't.

Yes. Someone's gotta create new grist for the LLMs so they don't Habsburg themselves.

doesnt matter if you use AI as long as you check the code afterword and know what you are doing

Does it make sense to still walk? I mean, we have cars. You want to go somewhere that's 1km away from home. Does it make sense to walk?

Someone wants to give you a phone number. Does it make sense to still try to memorize it or must you hunt down paper and pen to write it down?

... you get the drift? Do what makes sense to you.

When your learning it does and in work no

I'd love to vibe code with an army of agents, I just don't know if I could afford to vibe debug the resulting mess lol

Everyone thinks they’re an expert, but AI is brand new and nobody knows what the fuck they’re doing or what the ramifications are going to be. I see a lot of people cheerfully deskilling. I think it’s going to end up being a costly mistake to employers and employees alike.