Programming Deflation
tidyfirst.substack.com
tidyfirst.substack.com
Not saying you should disregard today's AI advancements, I think some level of preparedness is a necessity, but to go all in on the idea that deep learning will power us to true AGI is a gamble. We've dumped billions of dollars and countless hours of research into developing a cancer cure for decades but we still don't have a cure.
To put it in your words, I don't think LLMs get us 90% of the way there because they get us almost 100% of the way there sometimes, and other times less than 0%.
So yeah, very different.
You've just moved the problem from "I can't solve this" to "I can't trust if the LLM solved this properly"
Less cynically, it doesn't matter whether some code was written by a human or an LLM -- it still has to be tested and accepted. That responsibility ultimately must end up on a human's desk.
And, further, the existence of offshore, low quality, contractors that are in such frequent use.
AI replacing outsourcing and (sadly) junior SWEs is more likely than it just eliminating coding jobs across the board. Lord help them when our generation of senior SWEs retires, though!
Note: That isn't my retirement plan, but it'll just be a fun source of extra money if I'm right.
Even before remote work, you could easily pull a mid level developer with 3-5 years of experience who is probably underpaid for only 20% more than a junior dev.
I will be living in Costa Rica somewhere retired by the time that is an industry problem best case or be able to command a higher salary best case.
Then we explored AI tools and recently we adopted BotGauge. Does a pretty good job in terms of regression suite and much better decision than offshore low quality teams
The selling point is that it unreliably solves all your problems, "this model is PHD level intelligence!", compilers we know solves some problems and can't solve others, for LLM what they can actually do is very poorly understood by most since marketing oversells it so hard.
I know in the very right circumstances the "all-in" way could be faster, but it's already significant that I can do 5x as many coding tasks or do one in a fifth the time. Even if it never ever improves at all.
"Augmented programming" can be used to refer to a fully-general-purpose tool that one always programs with/through, akin in its ubiquity to the choice to use an IDE or a high-level language. And in that sense, I think your analogies make sense.
But "augmented programming" can also be used to refer to use of LLMs under constrained problem domains, where the problem already can be 100% solved with current technology. Your analogies fall apart here.
A better analogy that covers both of these cases, might be something like grid-scale power storage. We don't have any fully-general grid-scale power storage technologies that we could e.g. stick in front of every individual windmill or solar farm, regardless of context. But we do have domain-constrained grid-scale power storage technologies that work today to buffer power in specific contexts. Pumped hydroelectric storage is slow and huge and only really reasonable in terms of CapEx in places you're free to convert an existing hilltop into a reservoir, but provides tons of capacity where it can be deployed; battery-storage power stations are far too high-OpEx to scale to meet full grid needs, but work great for demand smoothing to loosen the design ramp-rate tolerances for upstream power stations built after the battery-storage station is in place; etc. Each thing has trade-offs that make it inapplicable to general use, but perfect for certain uses.
I would argue that "augmented programming" is in exactly that position: not something you expect to be using 100% of the time you're programming; but something where there are already very specific problems that are constrained-enough that we can design agentive systems that have been empirically observed to solve those problems 100% of the time.
That is very inaccurate. There are still a million problems to solve. nuclear waste is extremely dangerous. Even "normal" trash has been proven dangerous and difficult to get rid of with landfills leaking dangerous materials into ground water.
Accumulating nuclear waste for millennia hopping that nothing goes wrong is far from a proven solution. People cannot trust a solution when all concerns for people safety are disregarded.
> Economics gives us two contradictory answers simultaneously.
> Substitution. The substitution effect says we'll need fewer programmers—machines are replacing human labor.
> Jevons’. Jevons’ paradox predicts that when something becomes cheaper, demand increases as the cheaper good is economically viable in a wider variety of cases.
The answer is a little more nuanced. Assuming the above, the economy will demand fewer programmers for the previous set of demanded programs.
However. The set of demanded programs will likely evolve. So to over-simplify it absurdly: if before we needed 10 programmers to write different fibonacci generators, now we'll need 1 to write those and 9 to write more complicated stuff.
Additionally, the total number of people doing "programming" may go up or down.
My intuition is that the total number will increase but that the programs we write will be substantially different.
Or simpler :) I'd argue that in the past we needed more programmers for more complicated stuff (more hand-rolled databases, auth solutions etc. - a lot stuff was reinvented in each company), now we need many more people to glue some libraries and external solutions together.
The future could look similar, a lot of LLM vibe coders and a handful of specialized fixers.
Who knows though. Real life has a lot of inertia. One will probably do just fine writing just enterprise Java or React (or both!) for the next 30 years. I plan to be dead or retired in the next 30 years.
The Keynesian bogeyman of the deflationary spiral ignores intertemporal effects. Cell phones and laptops are getting cheaper all the time, but no one drops into an infinite wait because of time-preference. In the context of producing software becoming cheaper, people at a definite point value having a usable system today over a marginally cheaper version tomorrow.
People learn new skills out of interest or opportunity. A lot of people working on Windows native apps started working on web or mobile apps.
If we're going to go with economic/strategy models, I think the Laffer Curve is more relevant. Seriously extrapolating here: AI is optimal for many tasks which if used in those contexts can maximize productivity. Over-using it on unsuitable tasks destroys productivity.
Applied to AI I think it would be something like - ease of development increases the complexity attempted.
I think "How can this code be made simpler and any complexity either isolated or eliminated (preferably eliminated)?" should be the ensuing prompt after we generate things.
By making things simpler and/or more robust, you make some very complex algorithms much more feasible. And you end up with algorithms such as HTTPS or even raft being part of everyday life, despite their complexity.
I feel this is a bit like the "don't be poor" advice (I'm being a little mean here maybe, but not too much). Sure, focus on improving understanding & judgement - I don't think anybody really disagrees that having good judgement is a valuable skill, but how do you improve that? That's a lot trickier to answer, and that's the part where most people struggle. We all intuitively understand that good judgement is valuable, but that doesn't make it any easier to make good judgements.
I think the current state of AI is absolutely abysmal, borderline harmful for junior inexperienced devs who will get led down a rabbit hole they cannot recognize. But for someone who really knows what they are doing it has been transformative.
Do it every day for years.
Judgment is a skill improved through reps. Sturgeon’s law (ninety percent of everything is crap) combined with vibe code spewage will create lots of volume quickly. What this does not accelerate is the process of learning from how bad choices ripple through the system lifecycle.
Additional code is additional complexity, "cheap" code is cheap complexity. The decreasing cost of code is comparable to the decreasing costs of chainsaws, table saws, or high powered lasers. If you are a power user of these things then having them cheaply available is great. If you don't know what you're doing, then you may be exposing yourself to more risk than reward by having easier access to them. You could accidentally create an important piece of infrastructure for your business that gives the wrong answers, or requires expensive software engineers to come in and fix. You accidentally cost yourself more in time dealing with the complexity you created than the automation ever brought in benefit.
I believe the reason for this is that we still need judgement to do those tasks, AIs are not perfect at it and they spit a lot of extra code and complexity at times. Then now you need to reduce that complexity. But to reduce it, you need to understand the code in the first place. Now you cut here and there, you find a bug, but you are diving in code you do not understand fully yet.
So the human cognition has to go on par with what the AI is doing.
What ended up happening to me (not all the time, for example this for one-off scripts or small scripts is irrelevant, or to author a well-known algorithm that is short enough without bugs) is that I have a sense of speed that ends up not being really true once you have to complete the task as a whole.
On top of that, you tend to lose more context if you generate a lot of code with AI, as a human, and the judgement must be yours anyway. At least, until AIs get really brilliant at it.
They are good at other things. For example, I think they do decently well at reviewing code and finding potential improvements. Bc if they say bullsh*t, as any of us could say in a review, you just go ahead to the next comment and you can always find something valuable from there.
Same for "combinatoric thinking". But for tasks they need more "surgery" and precision, I do not think they are particularly good, but just that they make you feel like they are particularly good, but when you have to deal with the whole task, you notice this is not the case.
But time and time again programmers continue to demonstrate we have no self respect and we're happy to dance to the tune of capital for money. That means we're definitely going to be downgraded to "slop janitors" in the near future
More and more people are producers of information nowadays. But I think everyone can agree, the quality has declined substantially.
I personally find myself reverting back to textbooks more and more and trusting the information I find on the internet less.
I'm somewhat neurodivergent and got into tech precisely because it was a career where hyperfocus, compulsive systems building, passion for finding difficult solutions, etc. where valued. However, now it feels like those skills are no longer values; or even liabilities. As the article points out, companies now want me to "embrace commodity" and focus on plumbing code. However, those are precisely the areas that I'm not good at.
I really feel for the newer folks coming in. Writing software is pretty much all I've ever wanted to do, and I've been fortunate enough to make a decent career of it. But our jobs, like every other one, will be automated away sooner or later, and the astounding pace of progress in AI and robotics indicates rather "sooner".
The real question is what happens when there is just no more need for humans to work jobs. What are they going to do? Because tech trillionaires sure aren't going to provide a universal basic income for everyone out of the goodness of their hearts.
> We're not just <A> ... We're <B>
Is this proof of LLM writing or are people subconsciously picking up LLM patterns?
There's some other tells too, but we don't want to give the game away and let these people keep the ruse going.
However, there are other aspects to programming that can't be quantified, subjective components that are stripped away when delegating coding to machines.
The first most immediate effect I think is loss of the sense of ownership with code. The second which takes a bit of time to sink in and is at first buried by the excitement of making something work that is beyond your technical capability is enjoyment.
You take both of these out, you create what I could only describe as soul-less code.
The impact of soul-less code is not obvious, not measurable but I'd argue quite real. We will need time to see these effects in practice. Will companies that go all-in on machine generates code have the upper hand, or those that value traditional approaches more?
One thing I'm curious about is if those companies that go all in her to the state where they have the source but they now have vendor lock in with the ai vendor. Since no dev understands the code anymore
It's really hard to get an LLM to assist you when you don't know the right questions to ask. If your vision and convictions about the world are not strong enough, one plausible hallucination can take you into Narnia for an entire week.
Programming will still exist, it will be just different. Programming has changed a lot of times before as well. I don't think this time is different.
If programming became suddenly too easy to iterate upon, people would be building new competitors to SAP, Salesforce, Shopify and other solutions overnight, but you rarely see any good competitor coming around.
The necessary involvement behind understanding your customers needs, iterating on it between product and tech is not to be underestimated. AI doesn't help with that at all, at maximum is a marginal iteration improvement.
Knowing what to build has been for a long time the real challenge.
>We're not just experiencing technological change. We're watching the basic economics of software development transform in real time.
and I knew.
That's just the usual behavior of any market. When prices go down, demand goes up.
Jevon's paradox specifically says when prices go down by a factor of x, demand goes up by a factor greater than x, leading to total consumption (measured in money spent) goes up.
In the context of this post we can paraphrase (replacing money with programmers) like so:
Once AI makes it so developing any piece of software takes half as many programmers, the amount of software developed will go up by more than two times, leading to a net increase in the number of programmers required.
I also believe the difference between expert level and amateur programmers will be a lot easier to spot. Especially around the software or system architecture
Due to the inherent GIGO behavior of any software, it'll be a lot easier to tell that the code or software wasn't written by someone with deep knowledge since LLMs' output is just a good reflection of your prompts and creativity while at it(prompting). For instance, a prompt from Ken Thompson can't be exactly similar to your average Joe programmer. I've seen it a lot with most new open-source projects nowadays.
How would one even market oneself in a world where this is what is most valued?
Question 2: Do you think this will ever become valuable?
That's basically the job description of any senior software development role, at least at any place I've worked. As a senior pumping out straightforward features takes a backseat to problem analysis and architectural decisions, including being able to describe tradeoffs and how they impact the business.
"Every small business becomes a software company. Every individual becomes a developer. The cost of "what if we tried..." approaches zero.
Publishing was expensive in 1995, exclusive. Then it became free. Did we get less publishing? Quite the opposite. We got an explosion of content, most of it terrible, some of it revolutionary."
If it only were the same and so simple.
Maybe a few of them will pursue it further, but most won't. People don't like hard labor or higher-level planning.
Long term, software engineering will have to be more tightly regulated like the rest of engineering.
I'm also thinking about a world where more programmers are trying to enter the workforce self-taught using AI. The current world is the continued lowering of education standards and political climate against universities.
The answer to all of the above from the perspective of who don't know or really care about the details may be to cut the knot and impose regulation.
Delegate the details to auditors with AI. We're kinda already doing this on the cybersecurity front. Think about all the ads you see nowadays for earning your "cybersecurity certification" from an online-only university. Those jobs are real and people are hiring, but the expertise is still lacking because there aren't clearer guidelines yet.
With the current technology and generations of people we have, how else but AI can you translate NIST requirements, vulnerability reports, and other docs that don't even exist yet but soon will into pointing someone who doesn't really know how to code towards a line of code they can investigate? The tools we have right now like SAST and DAST are full of false positives and non-devs are stumped how to assess them.
Even before all this latest round of AI stuff it's been a concern that we overwork and overtrust devs. Principle of least privilege isn't really enough and is often violated in any scenario that isn't the usual day-to-day work.
That's for sure. Software is too important for not only the economy but for people's safety that it needs to be regulated. Adding ads to all software is making it fragile and prone to hacking. Bloating it with features to sell stuff instead of doing its primary function is equally bad for everybody. Don't let me start on bad engineering practices.
One limit on wages is $ of value / hour. If AI makes existing programmers more efficient, then you would expect total wages to go up.
If AI makes it easier for folks to become programmers, then the value produced could be split over more people. Alternatively, if you need fewer programmers then more value could be captured by a few superstar winners.
Only for economists who ignore the bulk of people and only focus on those who finance their luxury with loans.
What kind of also ignores the industrial revolution and all the progress since then.
I don't know this is indictment of Beck or economics, or both...
I suspect the reality around programming will be the same - a chasm between perception and reality around the cost.
Likewise, as lighting takes less electricity and the devices last longer on average people use more electric lighting over time.
Something similar might be happening in software. LLMs allow us to produce more software, faster and cheaper, than companies can realistically absorb. In the short term this looks amazing: there’s always some backlog of features and technical debt to address, so everyone’s happy.
But a year or two from now, we may reach saturation. Businesses won’t be able to use or even need all the software we’re capable of producing. At that point, wages may fall, unemployment among engineers may grow, and some companies could collapse.
In other words, the bottleneck in software production is shifting from labor capacity to market absorption. And that could trigger something very much like an overproduction crisis. Only this time, not for physical goods, but for code.
Is that capability a problem? We don't tend to do this unless the state subsidises things or labour unions protect things that don't have customers.
Not if you believe most other articles related to AI posted here including the one from today (from Singularity is Nearer).
I saw this happening way before LLMs came on the scene in 2015. Back then, I was an ordinary journeyman enterprise developer who had spent the last 7 years both surviving the shit show of the 2008 recession and getting to the other side of being an “expert beginner” after staying at my second job out of college for 9 years until 2008.
I saw that as an enterprise dev in a second tier tech city, no matter what I learned well - mobile, web, “full stack development”, or even “cloud”, they were all commodities that anyone could learn “well enough@ so I wouldn’t command a premium and I was going to plateau at around $150-$160K and it was going to be hard to stand out.
I did start focusing on just what the author said and took a chance on leaving a full time salaried job the next year for a contract to perm opportunities that would give me the chance to lead a major initiative by a then new director of a company [1].
I didn’t learn until 5 years later at BigTech about promotions were about “scope”, “impact” and “dealing with ambiguity”.
https://www.levels.fyi/blog/swe-level-framework.html
I had never had a job before with real documented leveling guidelines.
Long story short, left there and led the architecture of B2B startup and then a job working (full time) in the cloud consulting department of AWS fell into my lap.
After leaving AWS in 2023, I found out how prescient I was, the regular old enterprise dev jobs I was being offered even as a “senior” or “architect” were still topping out at around $160K-$175K and hadn’t kept up with inflation. I have friends who are making around that much with 20 years of experience in Atlanta.
Luckily, I was able to quickly get a job as a staff consultant at a third party consulting company. But I did have to spend 5 years honing my soft skills to get here. I still do some coding. But that’s not my value proposition.
[1] Thanks to having a wife in the school system part time with good benefits I could go from full time to contract to permanent in 2016.
I see it as a correction rather than deflation. We proved the shitty salesman of the 70s and 80s right. The computer really can be a bicycle for the mind; whether or not that bicycle actually fucking goes anywhere is still left to human wills.