This reminds me of when I rediscovered the track-listing of a CD titled "The Best Album of the Next Century Ever" which came with boxes of Golden/Cinnamon Grahams cereal. Had some real bangers like Josephine by Terrorvision and Come On Spring by Antenna, plus some very early tracks by Space and Muse who both went on to make it pretty big. I listened to it on repeat for months as a young teen. That kind of nostalgia probably isn't available to the generations below.
Today if they did that, there would be a QR code that you scan and then have to create an account to hear the music streamed from their website (which is really just a big ad), and then 8 weeks later when the promotion ends, the best album of the next century would be gone.
Hah! Come On Spring by Antenna was on the same, and only, album by Antenna that one of my white whales was on: 10-4 (explained elsewhere in the thread).
My guess would be that these particular problems were vulnerable to an attack which built on recent advances and potentially tied in something unexpected from a distant area of mathematics. "Harder" is becoming harder to define. Harder for humans is probably not harder for LLMs.
Attempting a serious but not-a-certified-whatever answer: "Points" do have meaning when properly used as a kind of moving-average tool for forecasting within a particular context.
Problems arise when people try to perma-peg them to particular tasks, or (worse) man-hours or (much worse) man-hours across teams. Even just encouraging the humans to answer in terms of hours/days taints the accuracy of the forecast by introducing a kind of bias.
My wife, despite loving all things French, just doesn't "do metric". She wants my height in feet and inches, my weight in pounds, boom done. So while I know my mass in kilograms, she needs the conversion done before she can even begin to have a reference point.
Upper management is the same way. They have forecasts that they need to make, deadlines and budget goals that they need to hit. They only deal in the units of hours and dollars (or local currency). Every software engineer is accountable for their work in those units only. The conversion needs to be done before the management chain has a reference point.
One easy way to do this is to have each engineer estimate the time it takes to fulfill a story after it's been pointed; then, upon completion, record their actual hours spent. Their estimated vs. actuals tend to stabilize over time, so even if they misestimate a task, you can arrive at a good guess at the time it will actually take.
> > Problems arise when people try to perma-peg [points to man-hours]
> Upper management [...] only deal in the units of hours
I feel you're mixing up different operations here. You can always express unfinished work as likely to require a certain number of team-sprints, which are convertible to theoretic man-hours. The key is that the conversation rate is only valid for a moment, and technically that moment was the prior sprint.
That's very different from management thinking (or worse, declaring) that points have a permanently fixed proportion to man-hours.
> [...] and dollars
If your management deals in international currency, then perhaps that would be a useful analogy to them: Points and Man-Hours are different sides of FOREX, and they fluctuate based on different conditions.
When Engineering predicts a group of tasks is 54 points, that's like a foreign company signing a long-term contract in €100 EUR instead of USD. You can estimate that you'll receive ~$112 USD in a year, but the actual dollars will likely be different because the exchange-rate will continue changing before that happens.
So, what you do is you recognize every ticket has a somewhat variable “actual effort”; and, if you’ve been honest in approximate effort pointing, you’ll know your team (or your own) velocity.
From there you can run Monte Carlo simulations - say a few hundred thousand, and get a pretty good estimate of actual time spent.
Assume that for various practical reasons, we've decided it's one of the best functions out there. How do we use it effectively, especially when it has noise, and drifts over time with unseen changes to the human_estimator and the hideously complex world_state?
A popular option is to run it multiple times with different person/task combinations, putting a projected number on to each task. Afterwards, the tasks finished in sampling period ("sprint") become a quantifiable total for that period ("velocity").
Do the same process again with the next set of tasks, and you can figure out which ones are likely to fit if the velocity doesn't change much. If you know the velocity will change due to losing staff or vacation days... well, we apply a multiplier and hope for the best.
Trying to "fix" the meaning of points is maladaptive, because they reflect many changing things which are outside our control and can't be independently measured.
Honesty isn't the issue, in my experience. The issue is unknowability. You usually can't know what clients actually want until you've given them a prototype and gotten their feedback. You often can't know the rough performance impact until you've done the first implementation. Sometimes you just have to work through a list of candidate solutions S1, S2, etc... and you can't know which one will eventually work. All of which means that forecasting beyond a couple of weeks is wasted effort, except for rote factory-line work.
I thought you weren't supposed to use it to estimate man-hours? Or is it ok if we don't allow engineers to provide time estimates, but permit simulations to do so?
Claude, vibe code me an entire startup, the actual product doesn't matter, but it should all be based on the incredible pun "turn 'sorry' points into story points."
/goal get accepted into Y Combinator, you have an unlimited token budget, be bold.
EDIT: no, do not just make a product that gives away your unlimited token budget to users for free!
Even if there's one dimension, this is not a logically correct statement. Some numbers are bigger than others, but there is no biggest number. Or it could be a partial ordering - say that a number x is "better" than y if y divides x. Then out of the first hundred numbers, some are better than others but there's no single "best".
> the water usage is heavily overblown compared to the agriculture industry
Data centers tend to concentrate their water usage within urban areas, causing a strain on local resources. The value of their output is much less tangible. Their water consumption scales up around tenfold if you consider the indirect usage from generating the electricity.
But on the other hand; yes, it would also help enormously if people ate less meat.
By “strain on local resources” are you just stating the fact that they use resources? Or are you trying to imply that it causes problems for those resources?
Companies don’t just buy land, connect to water and power, and then the utilities get caught off guard when the water system is suddenly strained by water going somewhere. There are processes for approving this before connection.
If a new project is causing a “strain on resources” then the utility planners have failed that community.
I agree that utility planners have their work cut out to plan for data centers. It's a different paradigm than what they're used to. I think they are being caught off-guard, but it's a matter of process and infrastructure. They typically plan around averaged or linear trends, but data centers can have volatile usage spikes, and supply can depend on e.g. seasonal rainfall which is challenging to predict.
Households with yards will have far more volatile water usage spikes related to seasonal rainfalls than a fully built out datacenter will.
A recent datacenter proposal near me was going to be built on a ~35 acre site. The entire site projected to use ~300 gallons of water a day when fully operational. That's including toilets, irrigation, etc. A household on a quarter acre lot in that area often uses 10-15,000 gallons of water in a month. 35 acres cut into quarter-acre-ish lots means a bit under 140 households. We'll cut it down and say 100 after streets and what not. That lot being a neighborhood like the others nearby would probably use at least a million gallons of water in a month, about 33,300+ gallons a day.
> A typical data center uses 300,000 gallons of water each day (equivalent to the demands of about 1,000 households), but large data centers can use an estimated 5 million gallons of water each day
Only if allowed to. Most data centres I've used have no connection to water from the data centre halls other than to the toilets. Using open loop water cooling is a cost cutting measure it's a choice whether or not to allow.
Firstly, who is deciding whether to "allow" open loop cooling? And secondly, from what I gather, closed-loop sites require an order of magnitude more power than open-loop to run. That's still going to have a large impact on local resources, especially as around 2/3 of power plants supplying data centers are also ocated in water-stressed regions [0].
> Firstly, who is deciding whether to "allow" open loop cooling?
Your government.
And I don't know why you put allow in quotes. This is a matter of laws and regulation, so it is used in its most literal meaning:
Forbid construction that does not account for the externalities imposed by its water and energy usage.
> And secondly, from what I gather, closed-loop sites require an order of magnitude more power than open-loop to run.
Are you suggesting that removing x kilowatts of input takes at least 10x killowatts?
Or are you suggesting they spend 10x more energy on the cooling than an open-loop system?
The latter might well be true - an open-loop system is after all trading water for energy in a very extreme way - if you have a water source with pressure the simplest water cooling system "just" routes water over radiators. In practice you'll want a bit more than that for most setups, but it can be very cheap if you can ignore the volume of water used.
That is still a regulatory issue. You can decide which resources are critical to you in a given location, and you restrict or allow different approaches accordingly, or require funding or licensing to suit.
> That's still going to have a large impact on local resources, especially as around 2/3 of power plants supplying data centers are also ocated in water-stressed regions
Still entirely a question of regulating externalities.
Not all power generation requires significant water use. Your government is free to e.g. require new data centre capacity to be accompanied by an according buildup of generating capacity with according restrictions on water use.
If that means data centres moves elsewhere, then so be it. That is part of the goal if you argue for pricing in and regulating externalities:
Ensure that the cost of operating in a way that doesn't negatively impact the community is priced in so that decisions are taken based on the actual full cost to society.
So to go full loop to my original comment (sorry): It's entirely a question of whether or not you want to allow it.
The government obviously has the power to regulate data centers. That's why people are petitioning them to do so, and why the lack of data transparency is frustrating. I'm not sure what argument you're making.
The point I made was that the water use is not a property of building a data centre, but of building a data centre without regulating water use. As such there are plenty of data centres that don't use much water.
That matters when considering what to regulate. A lot of people pushing for regulation are pushing for blocking data centres instead of regulating the externalities they claim is their issue.
I disagree. There is community opposition against individual data centers. There are calls for moratoriums until rules and regulations are improved. Calls for outright blocking data centers even with improved regulations are a small minority in my experience.
> If a new project is causing a “strain on resources” then the utility planners have failed that community.
Call it an echo chamber but when it comes to data centers that's the preponderance of news that I read. These two stories are about the county next door to mine. The first story highlights how the builders of the data-center continuously broke zoning laws. The second story from a month later describes how the board of commissioners let all that slide and approved construction to continue.
Those guys were fined a "record" $1.1 million, and:
> The state is mandating DataOne apply for and obtain the required permits or cease operations, but is giving the company 45 days to do so, according to a NJDEP spokesperson. During that time, regulators are allowing the facility to continue running the generators.
We're talking about an order of magnitude increase in the number of data centers in less than a decade and way more than an order of magnitude increase in the power needs because the AI data centers use a lot more power than even the big data centers of several years ago.
That's way beyond what utility planners can reasonably plan for.
> we demonstrate that data centers are an overwhelmingly urban phenomenon, with 97.5% of facilities situated in metropolitan and micropolitan statistical areas
“Metropolitan area” is doing a lot of work. Every metropolitan area I’ve lived or worked in has had industrial areas, too. Those industrial areas have buildings that use a lot of electricity, water, or natural gas and produce a lot of noise.
This isn’t some unusual outcome. It’s like one of those maps that purports to show one thing but really just shows population density.
Most data centres in Australia (even the announced biggest one in the Southern Hemisphere) are close to a water processing plant in an industrial zone in Sydney. The same plant that supplies 85% of Sydney with drinking water. At least there's an upgrade.
If the data center is built despite multiple demands to desist, then it may not be a surprise per se, but it can still be an unaccounted stress on local resources.
94% of the US population lives in a metropolitan or micropolitan area. There are ‘micropolitan’ areas in my state that’d I would colloquially consider rural, and there are large areas of the metropolitan area I live in that I would consider rural.
The Netherlands does not have areas sufficiently far removed from urban areas for this to matter. In fact the entire country has a managed water table, the bodies which govern it are the oldest demicratic institutions in the world (over a millenium), and survived any change in ruler country and government in the meantime.
There's no river or creek not used for drinkwater consumption or management, and they are all connected to agricultural use and transportation (canal water levels) as well. It's one big system, and farmers cheating on how much they pump up is already too big of a proboem for this to not matter.
In my part of the US there is enough rain that no extra water is used to grow alfalfa. The only water usage from beef production here is what the cows themselves drink.
> Data centers tend to concentrate their water usage within urban areas
Ahh, what? The world's greatest concentration of data centers is in the suburbs surrounding Washington, D.C., USA. Almost none of that would be classified as urban. I offer a counterpoint: The vast majority of data centers are suburban, rather than rural or urban.
> Their water consumption scales up around tenfold if you consider the indirect usage from generating the electricity.
> But on the other hand; yes, it would also help enormously if people ate less meat.
At first you had me, but then I realised it's not true if the price is building more data centres! That was just a trick to make the climate worse! Grr /s
Essentially, telling people to go vegan and get rid of their golf courses is going to piss them off, hence the reaction people have to being told that datacenters are more important than their comfort.
I'm personally willing to change my life to help the planet and nature in general to the point it doesn't affect my wellbeing (as in health and safety). For me, personally, the problem is not changing per se, but people finding no amount of change enough, always moving the goalposts with "but, what if" clauses.
The problem with datacenter water use is the state of the water leaving the data center. Agriculture and meat industry can reuse the water multiple times from cleaner to dirtier processes and return it to the nature with reasonable processing.
The additives used by datacenters in some cases can't be filtered or processed out practically or in economically viable ways, so the water returns back as unusable, esp. if they get water from a body of water and return it with added chemicals the process gets very damaging.
> The additives used by datacenters in some cases can't be filtered or processed out practically or in economically viable ways, so the water returns back as unusable, esp. if they get water from a body of water and return it with added chemicals the process gets very damaging.
Why would any water be returned?
Datacenters (and literally every industrial process that requires cooling) use water for evaporative cooling. The water evaporates, then returns as rain elsewhere.
The only water being "returned" I can think of would be from the intermittent or continuous blowdown systems to maintain TDS levels; but blowdown is a small fraction of the water consumed and is typically returned back via the city sewer system.
Roughly 80% of the water evaporates into the atmosphere. The remaining 20% is flushed out of the system to remove mineral build-ups and sent to local treatment plants.
It's complicated, and depends on the hardness of the water entering the system from the water supply as well, since blowdown is ultimately to control the level of dissolved solids in the water so it does not contrate enough to start coming out of solution. Rarely if ever is true demineralized water used in open loop cooling systems; it's simply too expensive to produce to then "throw away" into evaporation.
The previously stated 80/20 isn't a bad general estimate, but it varies depending on the specific site. 10%-25% of makeup water being lost to blowdown is a reasonable range.
As for papers about it, I was only able to find a few in my brief search, but they don't land on a single number since it depends on the ratio of dissolved minerals/solids in the circulating cooling-tower water compared with the incoming makeup water, which varies due to acceptability for the process and due to the incoming water quality.
This short PDF on "Mechanical Systems Water Efficiency Management" from the EPA goes over the concepts pretty well though
This document from the US DOD (UFM 3-230-13, Industrial Water Treatment Operation and Maintenance (2026), Chapter 5) says that most cooling towers operate at 3-10 CoC, with 5-7 CoC identified as a cost-effective range. Where CoC is Cycles of concentration described in the previous PDF, and a higher CoC means less blowdown and better water efficiency, with the trade-off of a greater risk of scaling, corrosion, and biological/water-chemistry problems.
We're talking about water that's released in general use, right? I would expect a closed cooling loop to have some biocides but have almost none of that leave the system.
And one datacenter had a weird bacteria that one time.
Blowdown is not clean water. According to UGA Extension (Saha, 2026) and EPA guidance on water-related permits for data centers, cooling tower blowdown may contain:
Concentrated salts and minerals — calcium, magnesium, silica, and other dissolved solids at 3–5× the concentration of incoming makeup water, depending on the cycles of concentration the system operates at
Corrosion and scale inhibitors — chemical additives applied to protect the cooling system's metal components from scale buildup and corrosion; typically phosphate-based or polymer compounds
Biocides — chemical treatments to control Legionella and other biological growth in the cooling tower; commonly chlorine, bromine, or isothiazolone compounds
Heavy metals — leached from piping, heat exchangers, and cooling system components; copper, zinc, and chromium are common depending on metallurgy
Glycols — in some configurations where freeze protection is required
PFAS compounds — some liquid coolants used in immersion and direct liquid cooling systems contain fluorinated compounds; some refrigerants used in air cooling are classified as F-gases (a PFAS subcategory)
That's a really bad source. I'm sure "may contain" is true but it gives no numbers or likelihoods for anything beyond the first item. And anyone with a reverse osmosis system is also purging extra salt/minerals. Not that extra minerals are even a big deal.
Heavy metals from pipe leaching can also happen anywhere.
And the explanation for PFAS is "what if another cooling system leaks into this one??". It's really a stretch.
The rest of the list might matter but again it needs numbers. If trace amounts count then home users need the same list; they'll put nearly anything down the drain once in a while.
One cubic metre of water is 1000 liters. A typical 25m swimming pool ("short course" / half of olympic size) is 20-25m wide. That's at least 500 m^2 surface area. At an average depth of only 1m (very conservative even for a recreational pool), that is 500,000 liters just to cycle the water once a year. Estimates I can find suggests 3-7 times as much as that on a yearly basis for a typical recreational pool.
So unless you're talking about a tiny private residential pool you're at least an order of magnitude off.
As someone who has been maintaining a pool for several years, you really hope you don't have to cycle the full amount of water once a year. You're really doing something wrong or incredibly unlucky if that's the case. That's at least true for the climate I live in. If I lived in a different climate, I'd probably forego having a pool.
It was incredibly painful when I had to drain the pool for deep maintenance a few years ago, and I'm not looking forward to having to do it again in a little while.
Also, most residential pool owners don't have half olympic pools in their yard. They're usually much smaller than that.
> Also, most residential pool owners don't have half olympic pools in their yard. They're usually much smaller than that.
Hence my caveat at the end. When talking about swimming pools, tiny residential pools is certainly not what springs to mind for me.
Most pools near me are in fitness centres etc. and are 25m pools.
And I certainly wouldn't go swimming in one of those if the water wasn't cycled regularly. Thankfully the data suggest they are.
If we're talking tiny little residential pools, then using a swimming pool worth of water isn't very shocking. Unless I'm way off, a 10-15m shower every day would use a similar amount of water in that case.
> I certainly wouldn't go swimming in one of the water wasn't cycled regularly
Once again speaking as someone who has maintained pools for a while, if the water was properly monitored and maintained I probably wouldn't mind. I'd absolutely take a pool that's actively properly maintained but hasn't had a full replacement in a few years over one that's poorly maintained but was fully drained and refilled a couple of months ago.
From experience while my son was swimming competitively, they had to drain most of the pools he swam in completely more than once a year because of various contamination, in addition to the ongoing cycling...
I can't imagine the state they'd be in after a year.
Once again, if its monitored and maintained properly, sure why not? Partial cycling when things that can't be filtered get too high (like TDS or CYA for an outdoor pool) but otherwise make sure sanitization is happening properly, pH and other metrics stay balanced, and good filtering is happening, I don't see a problem.
> more than once a year because of various contamination
Sounds like that falls into the "unlucky" category.
If its that you don't like thinking about tons of other bodies have touched that water you're swimming in, remember that dinosaurs pooped in much of the water you're drinking today.
> Partial cycling when things that can't be filtered get too high (like TDS or CYA for an outdoor pool)
Partial cycling in a high traffic environment seems like it is clearly ending up being a lot of water when every number I can find suggests it is generally expected to add up to several times the total pool volume per year.
> Sounds like that falls into the "unlucky" category.
To me it feels like it falls in the "expected" category somewhere that is dealing with hundreds of swimmers including small children every day.
Keeping in mind this was across dozens of pools - he swam competitively, not a one-off or rare occurrence in a pool or two.
It seems like we're mixing up terms here and miscommunicating. I thought we were talking about actually replacing the water here (as in, down the drain, fresh water from the tap), not just filtering as in recycling water through filtration systems. The total volume of the water should go through filters every few hours. I'm 100% not for the idea of only "partial filtering".
> To me it feels like it falls in the "expected" category somewhere that is dealing with hundreds of swimmers including small children every day.
Nah, that is definitely not normal to have to completely replace the water multiple times in a year because of water contamination events happening. The water can usually be sanitized. It might require a day's worth of hyperchloronation, but not often an entire drain and fill. That will happen often, probably several times a year. Usually only if your CYA is too high will it require you to actually drain a significant volume of water, which I agreed was one of the reasons to at least do a partial drain and really only applies to outdoor pools. Doing a hyperchloronation will likely increase your TDS, which won't go down until you drain those contaminates. But usually you can get those numbers back in an acceptable range without having to drain and fill the whole pool.
If the pool was properly maintained, what state do you expect the water to be in after a year without a total drain and fill? Do you have experience actually maintaining a pool?
> It seems like we're mixing up terms here and miscommunicating. I thought we were talking about actually replacing the water here (as in, down the drain, fresh water from the tap)
We are. I changed it while you were writing.
Everything I'm seeing suggests typical loss to backwash of the filters for a short course pool is in the 5k - 10k liters of water a day range, possibly less if you invest in expensive additional filters. 5k a day is ~1.8 million liters a year.
To get below a full replacement every year for a small short-course pool, you need to lose less than 1370 liters a day, or ~1.4 m^3
You're shifting things from normally needing to fully drain and fill the pool multiple times a year from specific contamination events to now talking about routine backwashing events, which I agree do happen as a part of normal routine maintenance (a part of the "partial drain" and "good filtering" I was talking about before).
I agree you'd probably use something like 5-10k liters of water for a backflush for filters that size, but that wouldn't normally be a daily thing. The amounts needed here would be highly variable depending on a wide range of factors.
I still stand by if you're having to actually fully drain and refill your pool multiple times a year (as in, an actual drain and fill, not slow partial replacement from weekly/every other week filter flushes) you're either very unlucky or you don't know what you're doing maintaining the pool.
This went into the weeds a little bit, but I've tried going over the figures again. I'm not sure if I'm missing something here or someone got some units wrong somewhere, but data centers seem to use far more water.
A medium size data center consumes 300k gallons of water per day [0]. That's over 400 million liters per year.
Even if we go with the highest estimate vidarh gave of 1.8 million liters per year for a full-size olympic size swimming pool, that's a tiny fraction of what a typical data center uses.
> A medium size data center consumes 300k gallons of water per day
Not always true. This can vary greatly based on datacenter design. A medium sized datacenter proposed near me was planned to only use 300 gallons a day for example.
Your source even points to this being highly variable, assuming you pay attention to their wording.
> Large data centers can consume up to
"Can consume", "up to". Not some hard truth that they will consume that much.
Fair point. If you divide the 163.7 billion gallons annually among the 5426 data centers, the figure does come out lower at ~30 million gallons or ~100 million liters annually. That's still far, far larger than the largest swimming pools.
And besides that: news sites love comparing water usage to things like bathtubs, the average household or swimming pools. The headline would likely be "datacenter uses 100 swimming pools of water per year!" and not "datacenter uses 6 million liters of water per year!"
The other problem is that stereotyping an entire social group simply alienates the members of that group. They perceive that society doesn't trust or value them, so in response they don't trust or value society. See e.g. racial profiling in the London Met's use of stop-and-search powers.
If your ultimate goal is harm-reduction, then stereotyping is irrational.
Unfortunately it seems most of the people who I've spoken to that advocate for prosecution-by-stereotype don't want to understand the nuance involved in causing groups to become disenfranchised. In the eyes of the people I've spoken to there is never any reason to commit a crime and if you associate with people who do then you're no better than the criminals.
Don't you think that isolating people in poor conditions amongst hardened criminals might make them more likely to commit crime when they get released?
Tbf opposing gay marriage is quite an extreme opinion in much of the developed world. I suspect Eich's position on the matter is tied to his Roman Catholic faith, so although he showed a certain level of remorse in 2014, he never actually came out in support of gay marriage. He then later spread misinformation about mask-wearing during COVID, including suggestions that people could die from wearing a mask.
Prop 8, the anti-gay-marriage proposition that Eich donated money to support, passed with 52% of the vote.
Eich never clarified his position (nor did he retract it nor apologize) so we have literally no idea what aspect of this he supported along with 52% of the population of California.
> Prop 8, the anti-gay-marriage proposition that Eich donated money to support, passed with 52% of the vote.
I'm sure you'll eventually have someone chime in here that the proposition was "misleading" and there were "disinformation campaigns" that resulted in this 52% figure, AKA "doesn't count, we know what the vote 'should have been'." People forget that Obama famously stated "marriage is between a man and a woman" in 2008 (was he lying to appeal to more voters? Possibly, but the fact is he said it). It was a very mainstream opinion at the time.
He very clearly was. He directed DOJ not to defend DOMA Section 3, and it fell in the courts, just like all the civil rights promoters wanted. If he hadn't been elected, it would have taken longer. That's why we forgive him for lying. We aren't stupid.
He also publicly identified as Christian, though he wrote that his parents and grandparents are nonbelievers.
First, I think you're gonna find it's not exactly a sub-single digit opinion today either. Second, the left tends to be extremely selective about who they forgive. Chick-fil A has done their apologies, stopped their donations, the whole song and dance and there's still a burning hatred for them. What's he got to gain at this point? People have made up their mind on him and all the performative apologies in the world won't get him back in anyone's good graces.
> First, I think you're gonna find it's not exactly a sub-single digit opinion today either.
I never claimed otherwise. For that matter, a double digit percentage of people still think that interracial marriage should be forbidden. I'm smart enough to be on the right side of that debate. Eich isn't. Are you?
> Chick-fil A has done their apologies
They never apologized, and they famously kept donating to anti-LGBT political charities after they said they stopped in order to get a restaurant approved in Chicago in 2012.
> What's he got to gain at this point?
Self respect. It's normal to apologize when you've wronged somebody. Not apologizing shows you either think you didn't cause harm or you are too prideful to admit you were wrong.
I don't think it's fair to call it "extreme". Small minority in some countries, minority in most and about half of the population in some is imo a fair assessment.
Even in USA it looks more like 70-30 if I am reading the polls correctly.
> I don't think it's fair to call it "extreme". Small minority in some countries
What does that matter? I would expect us to not care about a small minority doing something everyone else does. Not to actively legislate against their right to do it. If anything, this makes it more extreme.
The point is that it's not an extreme position when significant minority in most developed countries (small in some, about half in others) doesn't agree.
This was also majority position 2-3 decades ago. The world didn't change enough since them for mainstream majority moral positions back then to be called extreme today.
There's been a lot of talk about the purpose of code review recently. It makes sense in the face of AI. Heres a link that was submitted a little while ago: https://mathstodon.xyz/@mjd/115096720350507897
And in response I wrote a non-exhaustive checklist of things that a code review can look for:
- Does it functionally achieve what it sets out to (as per tacker issue or PR description)?
- Does it have extraneous code? Leftover debug prints, private API keys etc...
- Does it have any obvious defects? Memory leaks, un-handled edge cases, security flaws, obsolete API calls, etc...
- Could it be more understandable? Add/remove abstractions, better variable/method names, more/less functional etc...
- Is the style consistent with the codebase and/or style guidelines?
- Are there obvious performance improvements? Hashset instead of list, lazy evaluations, etc...
- Is it sufficiently well tested?
I think LLMs are okay at most of these, and worst at the first.
“Net new” is one it seems to be particularly bad at.
I don’t think I have ever even once seen an LLM solve a problem related to overengineering by simply removing the overengineering. They always choose to add more epicycles and further compound the complexity.
Some times I do, but no not really. If it's just a simple change of one or two lines I'll do it manually, but larger changes are much faster with AI.
I'll lay out a vague plan and let the AI fill in the blanks, mostly it'll get them right and where it doesn't I'll just tell it to do it differently and how. Works great for me.
The thing is, leadership relies on the people actually building the software to provide concrete, accurate feedback about cost. Without that they have no chance to do a decent cost/benefit analysis.
But AI has engendered a collapse in developers’ ability to actually do that. Those of us who are stuck on the vibecoding bandwagon have lost the comprehensive understanding of the systems under our care that we need to understand and explain the quality and maintenance implications of a change.
Worse, if you happen to lose your mind and suggest the initial development cost is anything more than ~zero, your friendly neighborhood Claude keener will publicly shame you for not having sufficient faith in the Glorious Agentic Future. Product leadership will then have no choice but to side with them, not necessarily because they agree, but because they, too, are aware that we’re still in the phase of the hype cycle where openly questioning said hype is a career-limiting move.
The "do we want this" question can also apply to functionality, not just cost/benefit. I've seen AI volunteer "features" that aren't actually useful or are actively harmful to user experience but it generates them because they do make sense in different contexts.
I think we're kind of missing a layer of testing, that should sit above unit and integration tests.
Something akin to "meta-tests", which are not about testing the code itself, but the approaches taken by the implementation - i.e. architecture, understandability, terseness, etc.
These tests would operate on the source code level, even when testing code for a compiled language.
> I think LLMs are okay at most of these, and worst at the first.
LLMs are worst at not realizing problems that I'd call "meta" problems. Here's one example to illustrate it:
I was allowed by my employer to work on a small project within the large collection of the projects which all constitute the product the company sells. Like a few dozens of other projects, it's written in Python. The company doesn't have any explicit policies about how Python projects have to be organized, it requires testing, linting, a CI code to package it etc, but the guidelines are very permissive. It just so happens that, beside the guidelines, there's a tradition: every other Python project in my company uses the typical Python bloatware, like masonry with a lot of insanity and mental flips going on in pyproject.toml, which is, in general, very typical for Python community at large.
My project used none of that. Instead, I wrote a ~100 lines setup.py file (no dependency on setuptools/distutils) that assembles the wheel and runs project maintenance tasks in the same way (interface-wise) things used to work decade or two ago (eg. "./setup.py test" if you want to run unit tests).
The AI reviewer didn't bat an eyelash. Found some typos in the comments, a problem with Base64 formatting, and generally OK'd the whole thing.
I knew I was on my way out. And I generally enjoy seeing people having a fit of rage when they know they are wrong (especially, together with many more like them), and scrambling for arguments that they know to be lies. I felt a little bit vindicated for the years of suffering I had to endure working with what might have been the dumbest and the most entitled manager I had in my life. :D
Anyways. My point is: the AI caught none of it. It was very happy with my approach to Python project management.
* * *
While my story is... more of an odd case, where this does have much wider implications is the AI-generated code. AI-generated code often fails to match these meta-requirements. I've seen AI reviewer OK'ing a PR containing AI-generated 10K loc Python file. I human would probably break after reading the first 1K lines. But AI doesn't get "tired", it just kept picking on typos in comments, criticizing short variables names etc. And there are other aspects in which AI-generated code is weird to humans in the ways that humans simply won't accept it, but AI reviewer would completely ignore as non-issue.
reply