In my own tiny way, I applied this logic when trying to be a contributor to an existing and popular open source project. I found that puzzling out poorly-written bug report tickets, ignored by the extremely busy experts on the project, I could find work to do.
I started by looking at well written tickets (to-the-point descriptions, minimal reproduction examples), but that method got me no work: some expert on the project would have a pull request for that kind of ticket in just a few hours, whereas I would need at least a week to figure out the root cause and another few days to craft a fix and a test.
So I started looking at ignored tickets, i.e., tickets that had been sitting around for weeks or months with no activity. Those often turned out to be very poorly written tickets with either very little information, or a huge wall of text with much business domain-specific info. If those tickets contained repro examples at all, they were often complex and very long, using external libraries I never heard of, and devoid of table schemas and example data. Sometimes I could get the author to provide more information, sometimes the author would not respond. I might have to infer schemas and make up my own data, try different things based on a stack trace, install libraries I would have preferred to not install, etc. I would work on trying to reproduce the problem for a few days, and at least a few times I struck gold. Then I would work on a self-contained, minimal reproduction case, followed by a week of sussing out the root cause. It was pretty time-consuming, but I was able to get a few fixes merged using that method, such that I was no longer a total stranger on the project (which helped me get other things merged into the code base).
Seems like you tore through some outdated Jira tickets, which is cool, but what was the impact of your work?
Hard doesn't mean impactful. In fact that's one of the biggest mistakes I see junior engineers make! But glad you got that learning experience, as you become more senior you'll definitely develop a better understanding of what drives actual impact.
Why are you being so condescending? A person wants to get into an open source project, so takes work he knows others won't finish before him, and you're talking to him like he's a noob (you have no clue how long ago this was). Hell, you don't even know if the tickets were outdated.
This is from my memory of listening to the full interview a few days ago, not just this snippet: I am not sure he is advocating that people should avoid AI, as long as they recognize its limitations and dangers, and he admits it can boost productivity.
His argument is specific to himself only:
"Many programmers come at programming at a much more business oriented perspective. Their goal is to make a specific thing and sell it ... or release it as a piece of free software. And the goal is just to have the software. They don't really care how they got there. So if you're coming at it from that perspective, whether you programmed it line by line, or whether you use AI, doesn't really make a difference.
...
But for me, I just like the act of programming. That's always been kind of why I did it. It's just kind of a happy accident that I lived in an era where I got paid well to do that. But that wasn't why I did it ... It's not an argument about quality. It's not an argument about productivity. It's about a fundamental love of the act of programming."
I am in the same boat as Casey. I like to program more than I like the finished product (and for me, that's just fine, since I don't work for anyone, so I am not screwing up any company's productivity).
I'm part of the old guard too, and I like the process itself, but I also like the result. I just think his arguments why to avoid AI sound like they're taken from an academic brochure, not like fact-based assessment grounded in relevant practical experience. Casey frames the choice as a binary: either you write every instruction yourself, or you lose the craft; he completely misses that there might be a third path where the craft is expressed at a higher level of abstraction. He stated that he likes to write assembler; but he also likes to write in higher-level programming languages; same for me, and I don't see a contradiction or a need for a binary decision. There were indeed arguments in the seventies and eighties that "real programmers don't use high-level programming languages"; at the time there were good reasons for this arguments because optimizing compilers were in their infancy; so it was a fact that true performance depended on human ingenuity. Now we are back at a similar discussion, just on a different level.
The discussion over work-from-home now seems like such a quaint topic from a different age (even though I was very much in favor of it when it was a hot topic).
>The supreme court ruled last year that prosecutors’ decision to allow a key witness to give testimony they knew to be false violated Glossip’s constitutional right to a fair trial.
Since stuff like this occasionally happens[1], and sometimes needs to be rectified long after a guilty verdict, I don't understand why any state would support the death penalty. Glossip is still alive only because there was a snafu related to the lethal drugs intended to kill him. Had that snafu been overcome, there would be no defendant to retry.
[1] "Occasionally" describes only the cases we know about, of course.
I wish I could go back in time and tell my 10 year old self to knock it off with the polish jokes (which were all the rage at that time, although I can remember only one now).
These jokes were for a reason. The Polish imigranta weren't exactly the cream of the crop. Even we, the Poles from Poland, would laugh at those Pollack from Greenpoint/Jackowo jokes.
I worried this blog post was going to pivot into a marketing pitch for some product, but no, it just describes the issue where the AI tool that generates your code probably won't document its reasons for the choices it makes. That documentation problem exists in the pre-AI era too, except that the reasons might exist in the heads of your co-workers and could possibly be teased out.
I know nothing about AI code generation (or about AI in general), but I wonder if you could include in your prompt a request that the AI describe the reasons for its choices and actually include those reasons as comments in the code.
The AI can't really describe its reasoning, though. It can only look at its context history and find a justification (which it will then present as reasoning). In my experience asking the model "why did you do that" carries substantial hallucination risk.
True, though I have found that forcing (I use an agent skill to do this) an LLM's agent to document the reasoning behind each "decision" it makes seems to lead to better decision-making. Or at least, more justifiable decisions (even if the justification is bad).
Not only can it not describe its reasoning, it can't "remember" if you ask it later; it can only observe what is. Nor can it be consistent; I've had it shift reasoning numerous times as the questioning continues, only to come full circle to its original statement while it apologizes profusely for being misleading.
The model will always be completing the story you start with it. There's no opinion to uncover because there's no experience that occurred. It's impossible to know where your influence ends and the model's factual basis begins.
I think the issue that majority of agentic developers don't understand their design fully. Those holes, perviously, were have to be closed by a person eventually, and inherently you gain understanding in the process.
This is not the case eny more. I never thought LLMs write bad solutions, but when you let it think for you you loose something important, understanding.
And when something brekas, some people consider this being a proof that the llm is the problem. And in a sense it is. But you are also to blame.
And that's exactly the issue with this all industry - in order to move faster you don't need to "type" faster. A person bashing his hands on the keyboard randomly at top velocity can get pretty high wpm.
In order to move master you need to _think_ faster, be _snappier_ and _sharper_ and most people aren't.
I like how one of my colleagues phrased it in a company meeting. He asked one of the c levels "for llm s to 10x me, I need to let them make all decitions and dictate intentions. By percentage, how much of the code you want me to _know_?"
The default for Claude today is "no comments". Here's a conversation I had with Claude about it recently:
---
Me: I've notice that Claude is deleting my comments and tends not to insert many of its own. I don't want to go too crazy, but can we modify CLAUDE.md to dial up the comments a little? Like, try not to delete comments that are still relevant, and add a few explanatory comments.
Claude: [... proposal for modifying CLAUDE.md ...]
Claude: Want me to apply this, tweak the wording, or place it differently? Note that your global ~/.claude/CLAUDE.md-style instructions (visible to me as system guidance) currently say "default to writing no comments" — the project-level rule above will override that for this repo, which is what you want.
Me: Where is this rule that says "default to writing no comments"? That seems crazy?
Claude: That text is part of the system prompt I'm given by Claude Code itself — it's not in any file in your project or in your ~/.claude/ config. It's baked into the harness's default instructions to the model.
Claude: Specifically, two lines from my system prompt:
Claude: Under "Doing tasks": "Default to writing no comments. Only add one when the WHY is non-obvious..."
Claude: Under "Text output": "In code: default to writing no comments. Never write multi-paragraph docstrings or multi-line comment blocks — one short line max."
Well the comments tend to be superfluous "whats" (describing the code itself) instead of the more helpful "whys." And they're almost never the most useful "why nots".
> And they're almost never the most useful "why nots".
Really? I find that Claude really likes to write "why nots" in comments when iterating on implementations and fixing bugs, to the extent that the comments grow into spot-logs of overly-specific documentation of what was tried and why it was scrapped.
IME the comments are not useful though. I'm not the biggest fan of AI code, but the codes still way higher quality than the "this is what this does" comments
Perfectly possible IME, it just requires more time. Even if you understand all the (important) parts of your code and validate it, Ai still helps with productivity a lot. It's just not as fast as blindly vibe coding - not at first at least. Agents on their own without checks, and superficial prompts without understanding, result in agents either introducing bugs, or if you have an appropriate test harness, painfully slow, in my experience. It requires skill to use an agent effectively.
But I don't relate to "it probably won't document its reasons"... I spend more time telling it that it has over-documented than under-documented.
It always wants to include the specific thing it is fixing in the API docs for methods it touches. No, this is not the place for a changelog of why this method is implemented the way it is! Describe the interface and the contract, these details are non sequiturs and they are a distraction.
Isn't this what good commit messages are intended to address? If a person has gone so far as to allow an agent to write all their code, they're likely having it commit on their behalf too.
Writing a skill / set of rules around what makes a good commit message would encourage the LLM to record it's reasoning (however much we truly consider it to be "reasoning").
By default, Claude is set to avoid comments. Eventually I got tired of it deleting my own explanatory comments and overrode the behavior with an addition to CLAUDE.md.
I think Claude is just behaving like all those programmers who make a VERY BIG DEAL about how much they HATE HATE HATE comments that might reiterate what the code does so will go to extreme measures to avoid them.
I stopped going to McDonald's (which I previously visited about once per month) mainly because they got very expensive, and the price does not match the quality of the food (and they also are not that fast anymore). If I am going to spend that much, I could spend a little more a go to a much nicer mom-and-pop place.
A secondary reason is that they are American. Although I am American, I am currently a resident of another country that is targeted by American tariffs, so I am trying to buy local as much a possible.
I stopped going because the McDonald's closest to me stopped serving water. The only way for employees to fill a cup with water is to use the sink, and that's not an option offered to customers. There's no way to buy or be served a water, not even a bottle of it.
There's a McDonald's near my home that I can order from if I'm craving garbage food quickly and don't feel well enough to leave the house, but they only get my order correct about 20% of the time. Another 20% of the time they make the wrong thing (e.g. the wrong kind of breakfast sandwich), and the remaining 60% of the time they forget to put half the order in (e.g. we ordered three of the Minecraft happy meal cube things a while back, plus an extra chicken sandwich, and we only got two of the cubes and no sandwich, plus we were missing two of the drinks for the meals).
The tariff issue is another reason not to patronize them, but at the same time if everyone in Canada stopped eating at McDonald's then McDonald's corporation would take a hit and thousands of Canadians would be immediately unemployed and thousands of Canadian suppliers of ingredients (beef, eggs, chicken, vegetables, etc) would lose a ton of business, so while I'd rather order from A&W for dozens of reasons I'm not outright boycotting American chains the way I am with American products.
I’ve actually switched from Just Eat to slightly more expensive Deliveroo because the latter one makes it extremely easy (at least in the UK) to report wrong deliveries and you get your money back in a matter of minutes. Also make sure to do the McD survey via the link on the receipt - that usually improves things for a while.
In Canada they have a CAD$5 McValue meal deal, so USD$3.67 for a McDouble, small fries & small drink. Do they not have similar deals in your jurisdiction?
For what it is worth, I live in Calgary and the McDonalds near me does not have deals like that. There is apparently a huge range in prices across McDonalds in a city, so there may be geographic limitations.
I don't go either, and the price is part of the reason. (I would go for the ice cream in summer, or for their cheap drinks promos).
US McValue meals (my local location, ymmv):
$6 for a McDouble, small Fries, 4 nuggets and small drink.
$5 for a McChicken, small fries, 4 nuggets and small drink.
$2.50 for a McDouble itself.
A typical burger + fries + drink at McDonalds where I am in the US is now about $20. You can get something much better quality (& larger size) at FiveGuys for same price, and even some nice quality restaurants have lunch specials that cost the same.
McD was never good, but when it was $10 it was still an OK occasional convenient lunch option. At $20 there is zero reason to go there.
It's also slow. It's not 'fast food' anymore. You order than wait for 15 or 20 minutes. I've waited 40 minutes at an airport McDs for food after ordering. The food is expensive, not very good, and slow. So yeah, I don't go to Mcdonalds anymore.
Actually, it's not. Speaking from experience. I'm a sample size of one, but that's still one better than your extrapolation from zero.
One of the local supermarket chains here in Denmark (Salling Group) even puts a star on the price tag for products of European origin.
For larger purchases, I'm doing research on the product anyway.
I still sometimes buy American at times; sometimes there's no avoiding it for certain items. But on the whole avoiding American goods isn't that hard, and doesn't require much effort.
Given that the US is the only country in recent memory whose politics have shifted from "pretty normal for a western nation" to "unpredictable rogue state", it's not as though the list of "countries to avoid" changes that often.
Countries like Russia, Iran, and China have been very consistent in their philosophies and actions; countries like France, the UK, and Japan have also been pretty consistent. The only real change lately is the US.
There was a period post-Brexit when I hadn't moved away from Ireland yet during which I also did my best to avoid UK produced goods too.
Now that was a lot harder though due to the UK still being in the single market at the time, and on top of that just how integrated supply chains between the north and the rest of Ireland are.
I'm cautiously optimistic that the UK is moving back toward sanity though.
> One of the local supermarket chains here in Denmark (Salling Group) even puts a star on the price tag for products of European origin.
Many store chains in Canada have also started putting maple leaf icons on price tags for Canadian-made products over the last year or two, after the US did whatever they've done. But it's harder to avoid US-made products here, because so much is imported and it's the only country we share a real land border with.
My US-based car has an "O"-like character in my license plate number, and honestly I could not figure out whether it was the letter O or a zero (even after searching the state's DMV site for information on how to distinguish one from the other). The character is very square, so maybe that means it's the letter O.
If the headline is more correct ("Because Police Cameras Cannot Tell the Difference Between a Zero and the Letter O") than the article content (which contradicts that claim), then I am alarmed. Otherwise, I am less alarmed.
I started by looking at well written tickets (to-the-point descriptions, minimal reproduction examples), but that method got me no work: some expert on the project would have a pull request for that kind of ticket in just a few hours, whereas I would need at least a week to figure out the root cause and another few days to craft a fix and a test.
So I started looking at ignored tickets, i.e., tickets that had been sitting around for weeks or months with no activity. Those often turned out to be very poorly written tickets with either very little information, or a huge wall of text with much business domain-specific info. If those tickets contained repro examples at all, they were often complex and very long, using external libraries I never heard of, and devoid of table schemas and example data. Sometimes I could get the author to provide more information, sometimes the author would not respond. I might have to infer schemas and make up my own data, try different things based on a stack trace, install libraries I would have preferred to not install, etc. I would work on trying to reproduce the problem for a few days, and at least a few times I struck gold. Then I would work on a self-contained, minimal reproduction case, followed by a week of sussing out the root cause. It was pretty time-consuming, but I was able to get a few fixes merged using that method, such that I was no longer a total stranger on the project (which helped me get other things merged into the code base).
reply