258
this post was submitted on 17 Jul 2026
258 points (92.4% liked)
Fuck AI
7560 readers
1283 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Yeah I think you’re right, when it comes to AI as a coding tool. I heard someone ask something like "where on the scale from flashlight to violin would you put LLM tools for coding". By that, they meant, if you use a flashlight, and it doesn’t work, you’re pretty sure it’s not your lack of experience with the tool - the tool is just broken. But if you play a violin and it sounds like shit, it’s probably because you’re not versed enough to play it (or at the very least you don’t know if it’s you or the violin who’s at fault), assuming you don’t play the violin.
If one finds LLMs unhelpful for coding, is it them not knowing how to play it, or is the tool inherently broken? It’s very arrogant to believe one is so knowledgeable about a tool, one which is in constant development too, that it is not them who is using it wrong, but rather the tool that is wrong.
There’s a plethora of other areas of just criticism of the technology, but I think how useful it is for coding is not one of them.
Aside: your "manmade" line breaks make your comment harder to read on a phone. I guess it’s a hard habit to break, but I would suggest you try. Most places people type text to be read by other these days, one should account for vastly different resolutions and aspect ratios (e.g. vertical & horisontal monitors). Adding in line breaks where it looks good on your screen is like writing text for yourself to read. That’s my personal opinion at least
Reading comprehension is at such a massive cratering low, you really gave criticism on writing style for prose copy pasted from an entirely separate medium written by a completely different author
Jesus Christ people
Take it easy. And don’t just take my one comment as anecdotal evidence of reading comprehension. I will say, first off, I wrote the above shortly after waking up - perhaps not my most analytical piece ever. Furthermore, English is not my first language (kinda not even second). And finally, I have dyslexia, so I literally have worse than average reading comprehension skills.
Also, from my device, it seems only the first 2 paragraphs are in quotation, and everything after "yes" has regular formatting, which gave me the impression of that being the commenters own words, and with the weird line breaks. (I’ll add I know people who have typed like that but have switched, hence I made the remark.)
All I want to say is you shouldn’t use my misunderstanding of the comment I replied to as any evidence of general reading comprehension. It shouldn’t be. And you seem overly affected by this.
The fact that you have to "prompt-engineer" means that the tool has a terrible interface.
In your flaslight analogy, there's an on/off button that doesn't work. You have to flip the flashlight, tap it twice, check the batteries and flip it twice again. Then it turns on.
The fact that LLM coding tools are basically micro-managing with a special twist "knowing how to prompt", makes it a shit tool for a lot of people. And they are justified in thinking so.
You can't hone a skill you have little control over. It's like a hammer that strikes a random area within a certain radius of a nail.
I both agree and disagree.
As for having little control, I feel like that happens in other disciplines too. The best archers need to account for the stochastic nature of wind patterns, and thus they do not always hit their intended target. But still, their accuracy is far greater than a novice. There is both skill and "luck" - if you will - involved in the process. You definitely can hone a skill you have little control over in the beginning, I believe. Maybe this is the analogy that should’ve been made.
I will say, "knowing how to prompt" feels like a weird skill, and I think the way we interface with these tools is kinda wacky. Regular text feels to fuzzy.
But it does help immensely, I find, to take every wrong agentic coding output as a learning experience. When I know the answer, and the agent failed, I ask myself why it didn’t fint it. Taking these opportunities has led me to be more proficient in the use of skill files, agent files, subagents, context window management, etc., all which have improved the agents output massively. I feel like a year ago, it was just prompting and copy-pasting (at least for me), but now, with tools like OpenCode, I can get a better looping effect with fewer errors.
I am curious about your findings
Don't forget we are talking about a tool.
A good tool is easily documentable, and then usable with said documentation. A good tool has perfect repeatability It is easy to predict the outcome of using a good tool.
Current LLMs have neither. For all the user knows, it's a black box that takes input and poops out output. You can only steer the output after some output has already been given.
I think we disagree on the nature of a tool. Or maybe I am thinking about more than a "tool", but not using the appropriate word for it.
Image riding a very fast horse. There is no perfect repeatability, it might veer off, or react to something you didn’t see, but a good rider would probably manage to arrive at the intended destination. I think I view it the same with agentic coding.
Now, I agree a horse is not a tool, and that’s why I mentioned that my language is perhaps a bit imprecise. (I guess you could even call it an agent.) But maybe an "instrument" or an "implement" fits, although that’s a bit cold wording for a horse in general, as they have other qualities, as being cute etc.. But I think you get my point.
I very much think LLMs can fall under the description of "useful implement for coding" today. But I do think it is a very poor knowledge retrieval implement. In fact I sucks for all these one-off questions, I my opinion.
I think the more interesting discussion is about how these AI companies consolidate power even more, how they steal data, how they pollute like crazy. But I think people who dismiss their use in coding perhaps view LLMs more like a flashlight and less like a very fast horse moving you from A to B.
Context engineering is a skill analogous to communicating a problem to fellow maintainers. This typically involves articulating the issue within an issue tracker, pull request, or comment using written text. Therefore, proficiency in "prompt engineering" is essentially a measure of one's articulacy.
So it's not a "useful tool for coding" then, if the quality of the output depends of one's ability to articulate in a way that tickles the LLM "the right way".
Also, no matter how well you articulate, the output is still somewhat random. You can produce two different sections of code with (essentially) the same prompt, which also just means that theres an amount of luck and percieved randomness in a tool.
Imagine you want a piece of software written but you're hiring it out, like a lot of corporations do. You write up a loose specification and hand it to two different dev shops and you'll get two wildly different results. The tighter the spec, the closer the results will be.
It's not much different from that.
Indeed. I agree that it's not a tool, but a type of outsourcing.
Following my analogy with the archer, promting the right way is like accounting for wind. And again, no matter how much the best archers in the world account for wind, they do sometimes miss. Still very impressive how high their skill is, even with randomness involved. And with agentic coding, one can always shoot twice.
TLDR: In short, like a lot of AI use the biggest problem is more how other people are using it that I have little control over.
The thing is that CodeGen utility varies greatly on what is being advocated and what situation it is being applied to. CodeGen getting things going when the operator isn't strongly opinionated about the details for fairly common patterns is fairly strongly in the wheelhouse. It is somewhat worse at amending a project in the same broad ways it was able to do when starting from scratch, even a project the model itself generated. So as you progress, it is more and more likely that a human will need to understand what is going on to be effective in modification. This presents a problem as people work themselves into a mess and can't get out of it and this causes a fair number of projects to just get abandoned because they can't go anywhere.
Even as it has demonstrable utility, the real world implications can be a mess. For example, the other week someone used agentic AI to open up 70 'security findings' on a project I work on. To it's credit, it found one actual issue, and while another issue was incorrect, looking into it I did find a separate issue it didn't notice, so I got two fixes out of it. However I had to deal with 68 completely stupid things that weren't anything. The operator at least up front sent an email that they didn't understand any of it and how much was real, but wanted to share in case any of it is useful. But I guess that's the price I pay for the two valid issues that might have otherwise not been caught. Then a few days after settling those, another user opened up over 60 and they were all dupes. Then the next week another user opened up a bunch that were all dupes too. Now it's ridiculous. One could argue that I could fight fire with fire and put an LLM on triaging the issues and closing out the dupes, but that means random github users can now make me spend my money on LLM services just by opening issues. Further, after being on the receiving end of AI chatbot triage in customer support, I hate inflicting that on the humans opening issues.
Then of course there's the code submissions and expectations around how I should handle them. There was a longstanding understanding that folks try their best, but implementing feature requests takes effort and folks are broadly understanding at delays or being a bit down the priority list. Now someone will Claude up a merge request instead of an issue and be impatient because "Claude already did the hard work, all you have to do is accept it", and it's a mess of code to review. It's not like the code at the hands of the operator is exactly good, for example someone sent a merge request because some feature raised an exception for them and blocked it from working and Claude "fixed" it. Problem is Claude changed the code to catch the exception, do nothing, and just say that it worked. In their test scenario, the feature was trying to make things the way they already were, so a no-op was no problem, but when I made their test case actually start from another state, it failed but still said it worked.
Another project I used to be a part of recently got handed over from the maintainers that long maintained it and lost issue to a super Slop-happy AI user. They got it rigged so it's all agentic and "addresses all issues and resolves all pull requests automatically from anyone". The project had some poor issue hygeine, so this meant the agentic code was doing things like fixing an issue from 2017 where the user complained it didn't work with Ubuntu 16.04. No idea if the fix actually worked, but it did make code changes that realistically no one would care about if they only did what they want. Except now they are getting bunch of new issues because they say "hey, the latest update broke a bunch of stuff I relied on", and another said "hey, it was great to see my issue get fixed last week, but this week the issue is back?" and messages saying "the update pace is ridiculous, why are there 3-4 releases a day, the former cadence was once a quarter". It's become a broiling mess of chaos because some AI enthusiast decided to make a nearly abandoned project that people were using a playground for AI usage. Dude even said point blank it's fine, just open an issue and AI will fix it for you and if it breaks, then someone else just open another issue and it'll get fixed and he is confident that eventually the users will shape the project into some equilibrium that way, so he has to do very little except pay for the tokens.