The AI Productivity fallacy
Posted by CriticalSink3555@reddit | ExperiencedDevs | View on Reddit | 64 comments
This article has been doing the rounds in my group and i'll be honest i'm pretty torn on it: https://readuncut.com/ai-and-the-productivity-fallacy/
It argues that if CEOs really believed in AI productivity gains, they'd be hiring aggressively to capture the surplus, and since they're laying off and buying back stock instead, the productivity story is just cover for cost-cutting.
imo the frame falls apart the moment you look at how mature software orgs behave when the marginal cost of output drops. They capture the surplus as margin or shift the labor composition, because in most of these businesses there is no uncaptured adjacent market waiting to absorb extra engineers. The author's own printing press analogy cuts against him, since most scribes did not get rehired as printers. The work expanded, the labor mix changed, a lot of scribes were out of a job. That is roughly what we're watching now, with junior SWE hiring down, senior hiring sticky, contractor spend up (or at least in my and my friends orgs), which is exactly what the HBS/BCG study he cites would predict.
The argument also assumes hiring behavior cleanly reveals what management believes about AI, ignoring the zirp hangover, higher rates, massive cloud capex commitments, and board pressure for margin after a decade of growth-at-all-costs. Companies that over-hired in the ZIRP years would be cutting now with or without AI existing.
however he does raise points that the productivity fallacy does not add up, sure it's 15-30% but this does just seem more hype than anything...again torn on this.
ForgotMyPassword17@reddit
My initial thought was your group circulated higher quality articles than most subs.
I do think the skill mix will change, but that’s happened before eg with most devs going from desktop apps to web
pydry@reddit
"We laid off workers coz of AI" is the only positive layoff story execs can tell.
If the reason is that growth is tapped out, thats a negative story -> stock sell signal
If the reason is management fucked up royally and hired all the wrong people -> sell signal + you've put a target on your back
Why on earth would management ever give a different reason if there were even the slightest shred of plausibility that it had something to do with AI (or the historic equivalent- automation)?
fcsar@reddit
Software quality is in decline
thank gosh!!!! (i’m an appsec engineer)
RespectableThug@reddit
Question for you: do you feel like you have better or worse job security than the software development engineers?
On the one hand, AI slop code must open up a ton of possibilities for exploitation. On the other hand, security is often one of the first things the bean counters like to take shortcuts on when money gets tight. Seems like a complicated position to be in.
fcsar@reddit
I’m not in the US or EU so my experience is pretty different. And I’m also working in a big international bank so we have tons of regulations that makes our jobs pretty stable.
That said, I’ve changed companies 3x since 2024 without much effort - while my SWE friends with similar YOE couldn’t even get interviews.
janyk@reddit
Or maybe your grandma is just a dope ass baller
The-Fox-Says@reddit
Grandma’s been spending most her life living in a gansta’s paradise
alexiskhb@reddit
it never occured to me that Automation could be a buzzword
Old-School8916@reddit
not just growth is tapped out, it was the end of a prolonged long era of zirp
pydry@reddit
well, the end of zirp was kinda what led to growth being inhibited.
if you wanna pull on the cause and effect thread all the way back ending zirp was an attempt to rein in a wave of inflation triggered by commodities spike driven by the ukraine war.
dipstickchojin@reddit
You mean "high interest rates" certainly
Old-School8916@reddit
true but dont forget COVID (both stimulus and supply chain disruptions). the economy was already running hot before ukraine lit a fire on it.
also COVID era hiring made it like a mini .com-like boom bust cycle
me_hq@reddit
On point
RespectableThug@reddit
I had that same Spotify issue myself a couple weeks ago. I barely listen to rap and my discover weekly was 100% gangster rap. SMH
CriticalSink3555@reddit (OP)
i'm not disagreeing with you i think that's a whole different thread, i'm just conflicted because i can see it on both sides i think my frustration boils down to whilst the article makes decent points it misses the biggest one for me which is a lot of companies don't have growth any more, it's just stagnant or minimal growth based on inflation price hikes each year and AI has given them a good excuse to suddenly become relevant again.
FlamingoVisible1947@reddit
If AI really made you more productive you wouldn't have to sacrifice quality in favor of speed.
I don't know why anyone needs anything else to realize that AI doesn't make software developers more productive.
simfgames@reddit
This is correct, you don't have to sacrifice quality for speed. You can get both if you know what you're doing.
po-handz3@reddit
Most employees use LLMs to get the same amount of work done faster, not to do more work. Story pointing hasn't really caught up.
You should see what's happening at the Founder/ pre-seed level though. Motivated people are moving incredibly fast
Empanatacion@reddit
He makes this same point several times and I don't think it's necessarily true. They're not just overseeing an army of keyboard monkeys that make cash by typing and now can type faster.
New business opportunities and new products have to be developed before you need an army to execute on them.
corny_horse@reddit
And that even if such opportunities exist, will the cost of this additional new X headcount be sustainable? Will it be sustainable if LLM prices double? Triple? 10x?
F1B3R0PT1C@reddit
My company has a history of making bugs and with our senior-most bug factories now armed with AI that number is going up exponentially because they’ve only gotten more careless
RespectableThug@reddit
This reminds me of a joke I heard a long time ago.
I had a problem with my code and my colleague recommended I solve it with design patterns. Now I have a ProblemFactory.
joseconsuervo@reddit
I like this joke
Mountain_Tax7379@reddit
how do you see ai affecting the hiring of junior devs in your company?
aeroverra@reddit
We only hire Indian contractors when it comes to juniors.
pmargam@reddit
It's crazy to invest in AI tools to create more PRs but not on AI tools that aid developers reviewing PRs more efficiently. We tend to be lousy at PR reviewing, they nitpick, we completely miss gnarly bugs due to fixating in fluff, etc. If you are struggling with the quality of the PRs being generated:
* Invest in AI tooling (CLAUDE.md/AGENTS.md, docs, etc.)
* Invest in AI reviewing tools (these days you can put something together very easily with claude managed agents, otherwise just use something off the shelf)
Expert-Reaction-7472@reddit
AI code reviews are almost too good now, spotting random edge cases that have existed forever. They obviously aren't enough of a problem to need fixing else they'd have been fixed, but it feels irresponsible leaving them once you know they are there!
CodelinesNL@reddit
The issie with AI is that since it removes a lot of the "writing code" bottlenecks, other bottlenecks surface. So now suddenly your QA gets overwhelmed and your product managers can't decide on features fast enough.
Those CEOs know this. That making development a lot faster doesn't mean everything else will too.
So IMHO that article misses the mark.
caprisunkraftfoods@reddit
This unfortunately isn't as crazy as it sounds, this is an endemic failure mode across the economy in general. Return is evaluated as a percentage rather than dollars, so anything that lowers the percentage gets passed up even it would ultimately return more dollars. As such we're suffering chronic underinvestment across the public and private spheres.
zacker150@reddit
Meanwhile, Antropic and VC-backed startups are actually hiring aggressively. The catch is that they're only willing to hire 10x engineers.
Adept-Log3535@reddit
This. My company and Anthropic are backed by same VCs and we hired 5000+ new full-time employees and doubled employee headcount since 2025. Our SWE interview process is brutal and the on-site passing rate is below 15%.
bzBetty@reddit
Anthropic must be taking all the other swe
gefahr@reddit
It sucks and it's going to be terrible for the industry long term, but it's what I'm doing as well. Only hiring top few percent engineers, the ones right below what Anthropic et al. are taking off the market.
I just don't have the headcount for even "decent" engineers, to say nothing of hiring juniors.
zacker150@reddit
Personally, I think this is a short term speed bump for the industry while we learn how to use these new tools.
Long term, I think that the junior role will return as best practices get solidified. Seniors and mid levels will write
PLAN.md, and juniors will steer & debug the agents doing the implementation.FriendOfEvergreens@reddit
Woah, hold on, you heard that companies should recruit from MIT? That’s crazy!
gefahr@reddit
I agree with that, but we're VC-backed and marginally profitable. I don't have the luxury of looking longer term right now.
Agree it's probably the right move for companies that can look beyond the next year or so..
gefahr@reddit
I agree with that, but we're VC-backed and marginally profitable. I don't have the luxury of looking longer term right now.
Agree it's probably the right move for companies that can look beyond the next year or so..
Tacos314@reddit
You're like the 100000th person to say this, it's not really new insight.
mechkbfan@reddit
Literally just need to use AI tools on enterprise code bases for a while to see this outcome
Like sometimes it's done incredible things for me that saved me hours e.g. Here's a prod exception & stack trace, go tell me the root cause
Let it run off in the background while I work on another task. Comes back and tells me what the issue was & fix. It was 80% correct but I reckon it still saved me an hour or two that I didn't have that day.
crazy0ne@reddit
I understand that it CAN be helpful, but at what cost?...
Save an hour here or there for several thousand tokens?
And there there is the reading overhead because everything generated needs to be scrutinize as we don't know when the 20% will hit.
lasooch@reddit
Try several million tokens.
Heffree@reddit
There’s a bunch of cost that’s hard to quantify as well. Hunting for that bug gives you an opportunity to reinforce understanding of the code base and pick up residual knowledge.
They probably had to reread most of the code path anyway.
mechkbfan@reddit
Agreed, it's a fine balance for me at the moment
For a period I tried to use AI first for everything, but I did see my brain felt like it was getting lazy/turning to mush, and it became harder to review it's code
Now I typically use it for mindless tasks, or initial investigations when curious and don't have capacity to look into it
e.g. recently just asked it to upgrade our Angular version. I'll still read the migration guide and what's new, but I have zero interest in doing the grunt work
mechkbfan@reddit
Admittedly I've paid no attention to tokens
If the concern is electricity/water usage, to me the proper fix is governments to introduce scaling costs of water consumption
i.e. data centers pay 10x the cost than house holds per litre/kw
And yes, there is the reading & scrutinisation overhead but I think the overall outcome is net positive
e.g. at the end of the day, when I'm tired, it's easier to review a suggestion that can easily be timeboxed to <1 hour than it is to kick off a new investigation that you have no idea how long it'll take
FWIW, zero care factor if AI stopped working overnight. It's just another tool for me that I'll use when it makes sense
wizzward0@reddit
Hate it when people try to discuss things on reddit!
YahenP@reddit
Whether there's AI or not doesn't matter. What matters is that the software market is a market with a limited and finite capacity, and it's long been 100% occupied. To increase turnover, you need to capture a larger market share by driving out competitors, not by increasing productivity. There's no other way. To increase profits (in real effectively remain at the same level, since competitors are doing the same), you need to reduce costs. And the first thing to fall under the knife is employees—that is, us. Increasing labor productivity in itself is meaningless in our market. It gives the business owner nothing.
LoaderD@reddit
Woah, you mean executive compensation optimizes for short term wins over long term growth?
Nobel prize type shit frfr ong
Famous-Test-4795@reddit
I would assume that when there was a lot more free money, it was easier for companies to have principles
OtaK_@reddit
There has never been more free money, as long as you agree to slap "AI" on your product. Look at OpenAI, no measurable profit or use, yet hundreds of billions have been poured into them.
It's just the "www" craze all over again, but actually stupid this time.
Foreign_Addition2844@reddit
Lmao What?
NeitherEchidna3491@reddit
Taking the "productivity increase is real" thesis at face value, most companies are just not cut out to suddenly pivot and leverage that productivity increase to make x% more features/products to translate into revenue. For pretty much anything but a startup cutting one of the largest cost centers vs investing in new things to increase profits is a completely rational & viable choice.
There are many more cynical & plausible takes about this as well but I don't think it is a fallacy personally.
KitchenTaste7229@reddit
Yeah, pretty common for companies to use AI as a buzzword to explain layoffs that are actually driven by macro factors. There's even a recent MIT-linked analysis on AI and worker adaptation that shows that overlap clearly, as layoffs attributed to AI often overlap heavily with cost-cutting, restructuring, and post-ZIRP corrections. Not really about purely automating workers. At the same time, execs are just incentivized to highlight ai productivity to justify huge capex and strategic bets. Ironic though that recently there's been more evidence that AI can speed up certain parts of work, but in general also creates more review, debugging, and oversight, which usually falls into the responsibilities of senior engineers.
ProfessorPhi@reddit
The actual answer is that silicon valley is capped in terms of product they can bring to market. I.e. even if they were more productive, there is nothing they can capture.
Most of their innovations were just advertising platforms, middlemen or ways to bypass legal restrictions. They've been capped out well before covid which is why the covid bump couldn't be maintained.
So they've overhired for the current market, and even if AI is productivity enhancing, they have no way to do anything new, but you can't tell since the layoffs would've come anyway. The bottleneck on SWE was rarely ever the SWE side of things.
fued@reddit
AI is just a nicer way of saying "offshoring"
CatDawgCatDawg2@reddit
What companies are offshoring?
Old-School8916@reddit
it's a relatively disrupted lukewarm economy in other places as well in terms of tech employment, tho it is certainly better than in the US
fued@reddit
I actually wonder what the tech news in countries where offshoring workers live is like, are they all having layoffs from "AI"?
True_Skin7151@reddit
I'm in India and it's AI layoffs here as well
fued@reddit
like is the market crashing? or just minor layoffs here and there
True_Skin7151@reddit
No hiring for juniors unless they are from highly reputed colleges. Seniors get calls rarely. There were major layoffs affecting Oracle, Amazon, etc. There are minor layoffs as well
thankyoujam@reddit
Yeah the article makes some good points, but I think is flawed in a couple of ways... as you mentioned it's not so easy to move to adjacent markets or to identify new opportunities for investment. It's usually easier to cut costs (or optimize something that is already in place) than it is to build something net new/different.
Honestly, in the companies I'm more familiar with, executives feel they need to invest in AI just to keep up with their competitors. Maybe they care about cutting costs, but it is usually because they think that competitors will be doing that too, and then could undercut pricing in the market.
It's a flavour of FOMO - "I better invest in AI productivity because if I don't our competitors _will_ and then they will surely leapfrog us."
Is it rational? Maybe -- you should probably do _something_ in this environment . But no one knows if what you do invest in will be helpful, hurtful, or neutral. It's honestly probably less likely to be neutral and more likely to be helpful/hurtful IMO. There's also a chance that no matter what you do, you're gonna be made irrelevant by this technological shift.
h____@reddit
I think both sides are right depending on where AI sits in your workflow. If it’s “generate code and pray,” productivity falls apart. If it’s used for bounded tasks (test scaffolds, migration scripts, repetitive refactors, first-pass docs) with human checkpoints, gains are real. The key metric I track is rework rate, not raw output. I shared my scorecard + guardrails on hboon.com in my coding-agent notes.
vansterdam_city@reddit
Your premise relies on the assumption that companies have a limitless supply of opportunities to apply that productivity. That is very company and team dependent.
At my own company things feel quite mature. I feel like we could easily cut 50% of our headcount and be totally fine.
Bricktop72@reddit
Or it means management only understands cost cutting. It's a safe play but they are leaving tons of value on the table.
Three-quarters of AI’s economic gains are being captured by just 20% of companies – with the leading companies focused on growth, not just productivity