Can we have a rule which says - "those who have not coded for a living shall not make sweeping statements about the future of coding" ?
All sorts of academic and management folks are coming out of the woodwork making predictions about the future of coding when they dont know what the hell it is ?
Academics are of no concern really. They aren't treated seriously, but I do care what management things. Even if they're blowhard retards with no idea what they're talking about, they're still calling the shots and still making the decisions
I gave co-pilot a shot this weekend. Building a toy app for doing a bunch of text processing.
Last time I tried, when this stuff was 'new' , I was thoroughly unimpressed.
This weekend, I did about 8 hours work in 2 hours. Copilot worked _almost flawlessly.
I cam away very impressed. I still wonder how well it would work against a large established repository where it isn't free to re-invent the wheel and doesn't know about established architecture and methods etc..
But it did change my perception overalll. They are getting better, and fast...
Are you trying to mix two payment models together though?
To provide a counterpoint:
With almost all traditional employment, people are paid for their time. Their output is more an expectation based on the time put in.
So if you want to bill by output, that's always an option as a consultant (fixed price).
But as a business owner, I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
If you were digging ditches by hand, but now you have an excavator with you, are you expecting 10x payment just because you can now do it 10x faster? Nobody will pay for that. Instead, you would now charge by a different metric (say, length of the ditch, rather than hours). And you would now be able to handle more customers since you can provide the same outcome to more people in less time.
I have no problem if someone wanted to charge by outcome. I think the concept of paying for "time" is going to go extinct in this new world. Think of lawyers who bill $500/hour. Now they can't say "I need 4 hours to have my team review this" and send you a $4k bill for their whole team to bill, because expectations have changed.
But framing this as an "us vs. them" thing by comparing apples and oranges isn't going to get us anywhere.
> With almost all traditional employment, people are paid for their time. Their output is more an expectation based on the time put in.
Yes and no. It's true that employees are rented for a limited amount of time rather than owned like tools. However, employees can earn vastly different sums of money for the same amount of time. Some make $7 per hour, while others (a privileged few) make $7000 per hour or more.
> I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
Ok, but are you going to let employees who use these "labor-saving" tools work 1/4 of the hours per week as before for the same compensation as before? I suspect not.
> But as a business owner
> But framing this as an "us vs. them" thing by comparing apples and oranges isn't going to get us anywhere.
Of course that's what a business owner would tend to say.
My point is that there's no inherent reason why the benefits of "increased productivity" would trickle down to employees. Individually, employees may look at A.I. and think it's great, but collectively, if all employees have access to the same tools, then employees don't benefit from A.I., because it doesn't fundamentally change their relationship with employers.
Think about how students started using A.I. to cheat and get ahead. This turned out to be short-sighted, because everyone has the same access to ChatGPT, and thus A.I.-fueled cheating became ubiquitous in school, and the cheaters are back where they started, no longer ahead of their peers and no better off than before. Indeed, they may be worse off now, because potential employers have to expect that all college graduates are incompetent cheaters who didn't learn anything.
This is not true at least with the current crop of AI.
I have seen 4-5 people use the tools (won’t say where and in what capacity, but only one of them was an employee of mine) and the results were wildly different.
One person just copy pasted AI responses and added no value. But that meant they also copy pasted substandard responses and it actually reduced the quality of their work (in fact, their greatest asset was client management)
Another one was a coder, and even though he was using windsurf every day, he was getting stuck on the most basic items. And having access to AI made no difference in his productivity because he doesn’t have the base knowledge and experience to be able to discern what is a quality suggestion and what isn’t. And he doesn’t know what questions to even ask the AI to help unblock him.
I actually think these tools will WIDEN the gap between what a good engineer can do, and what a mediocre one can. And it will widen income gaps. I’m not going to say this is a good thing. I’m still a human being.
For example, with regard to students, it was the mediocre, lazy students who started using ChatGPT to cheat. The good students didn't need it. And as I said, the cheaters are no better off than before. They mistakenly thought they had some completitive advantage, but the same tools were available to everyone.
With regard to engineers, the issue isn't the difference between good engineeres and mediocre engineers. The issue is that all good engineers have access to the same AI. Thus, engineers who were equal to each other before AI are still equal to each other. It's not a competitive advantage over peers, and thus there's no incentive to pay them more. AI doesn't change the supply of good engineers, nor does it give individual good engineers a competitive advantage over other good engineers. Any additional productivity comes more or less for free, because it's due to factors external to the engineer rather than internal to the engineer.
In the end, it's all supply and demand. That's not to say it has no effect either. Increased productivity from AI can increase wages by raising the upper bound for salaries (companies won't pay an employee more than the value he can create), and creating competition between companies (if a single programmer can generate $500k worth of value, and their salaries are only $100k, then we'd expect other companies to spring up to arbitrage this difference).
> companies won't pay an employee more than the value he can create
True, but the wealthy corporations aren't simply paying for "programmers". Cheap programming labor is available, especially in less wealthy countries, and corporations could avail themselves of that if they wanted (and they sometimes do). When some programmers make more than their peer average, it's because the employers consider those programmers to be "the best", the smartest, most skilled, hardest working, or whatever. In that regard, A.I. is largely irrelevant. The best programmers and the worst programmers can both use A.I., so it's not a distinguishing factor in the labor market. What matters the most is the supply of the type of programmers that companies are seeking.
A.I. might (or might not) increase a company's revenue, but there's no reason the company can't just book that as profit. A.I. doesn't increase the demand for programmers; if anything, it might reduce the demand for programmers, if the programmers they've already hired become more productive. Adding more employees has diminishing returns, as we've seen with mass tech layoffs in recent years. Hiring is not an automatic productivity boost, if you're measuring productivity as an addition to revenue. At the end of the day, producing more stuff doesn't matter unless the company can sell that stuff.
> if a single programmer can generate $500k worth of value
How do you even determine the monetary value produced by a single programmer in a large team effort?
(Ask yourself this: how much "value" is generated by human resources or administrative assistants or janitors? You might say $0, but nonetheless they get paid more than $0, because the work needs to be done. You can't a functional corporation without them. And their compensation is based on supply and demand, not on "value generated".)
Let me put it this way: I used to work for a small company, and some of my coworkers also worked at a BigCo at some point, either before or after SmallCo. Their personal productivity didn't change much over time, from one employer to another, but their employer's revenue per employee changed dramatically from one employer to another. A tech company's revenue depends on many different factors, many of which have nothing to do with the employed programmers.
Let me put it another way: the more that employers rely on publicly available automatated tools—for example, LLMs—rather than on the individual skills of employees to increase productivity, the more the employees themselves become interchangeable cogs in the machine, and thus by supply and demand, their compensation will not increase.
Of course, if using LLMs were an elite skill that required inborn talent and/or years of specialized training, then you would expect LLMs to increase the compensation of those with that unique skill, but I've seen no evidence that this is the case.
> But as a business owner, I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
s/these tools/drugs
And before you claim that it's an "apples to oranges" comparison, let it be known that we have no data on the long-term cognitive effects of offloading thought to machines.
Maybe because of supply and demand, but not because they bill hourly, which is exactly my point. It won't be simply because the excavator can get it done 4x faster.
I mean that the price multiplier isn't based on the productivity multiplier, as the parent was implying. Of course, the cost of the excavator is built into the price too.
Also for smaller jobs (like with coding), I might now be able to hire a mini excavator (using AI as an analogy for the excavator) myself for a weekend and just do the job myself, and leave the bigger jobs to the pros. That's already happening (look at RR Buildings on YouTube, two guys build massive structures with the help of tools).
> But as a business owner, I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
what this means is that the business owners capture all of the benefits of the increased productivity (which is what we've been seeing for decades)
You can keep seeing this through an adversarial lens. And I could just respond with some flippant remark like “well, they’re the ones that pay the bills” - but all that does is create a false battle between parties that have a shared interest (that of making a living and providing value)
Either way, I do care tremendously about the power of efficiency. That can sometimes lead to better profits, but more likely leads to happier customers and more business, which helps the employees.
>what this means is that the business owners capture all of the benefits of the increased productivity (which is what we've been seeing for decades)
Not really. Even though labor's share of GDP has steadily dropped for the past half century, it's on the order of 4% for the past 70 years. Real GDP per capita on the other hand has more than tripled in the same time period.
Because there's an enormous and growing gap between top compensation and lowest quantity pay, a better comparison would be to look at _median pay_ as a share of _GDP per capita_.
"business owners" by definition, means capital. AI engineers, quant traders, and bankers making 7 figures TC doesn't magically mean they're "business owners" in "business owners capture all of the benefits of the increased productivity".
>> still wonder how well it would work against a large established repository where it isn't free to re-invent the wheel and doesn't know about established architecture and methods etc..
You’re missing the point of AI coding if you think that’s the measure of its capabilities.
I admit, have a negative feeling that if my Computer Science discipline is going down in the Titanic, then other disciplines should go down with me. I know its not a good attitude. It feels unfair that CS job will be the first to fall, but we have had it so good for so long, maybe it is fair. Our industry moves so fast. I suspect lawyers, doctors, other "white collar" jobs will take longer and be slower to disrupt due to societal delays and protections (unions, laws, etc.), even though AI could potentially affect these disciplines more down the road.
So far AI seems to better in generating technical debt than value. Not exactly worried - in 3-4 years there will be a lot of investment in clearing up the currently created vibe mess.
AI is, currently, coming not for the coders who made it but for the coders who didn't contribute to or ignored it. The foundation labs are all quite committed to recursive self-improvement of coding tools as a general research accelerant.
When I see how unsupervised LLMs struggle with coding tasks (see public PRs from Copilot on Microsoft codebases) I don't see how recursive self-improvement can lead to any actual improvement rather than the total opposite.
Can we have a rule which says - "those who have not coded for a living shall not make sweeping statements about the future of coding" ?
All sorts of academic and management folks are coming out of the woodwork making predictions about the future of coding when they dont know what the hell it is ?
Academics are of no concern really. They aren't treated seriously, but I do care what management things. Even if they're blowhard retards with no idea what they're talking about, they're still calling the shots and still making the decisions
I gave co-pilot a shot this weekend. Building a toy app for doing a bunch of text processing.
Last time I tried, when this stuff was 'new' , I was thoroughly unimpressed.
This weekend, I did about 8 hours work in 2 hours. Copilot worked _almost flawlessly.
I cam away very impressed. I still wonder how well it would work against a large established repository where it isn't free to re-invent the wheel and doesn't know about established architecture and methods etc..
But it did change my perception overalll. They are getting better, and fast...
> This weekend, I did about 8 hours work in 2 hours.
How long until managers expect employees to do 4x as much work per week in the same amount of time, for no additional compensation?
Are you trying to mix two payment models together though?
To provide a counterpoint:
With almost all traditional employment, people are paid for their time. Their output is more an expectation based on the time put in.
So if you want to bill by output, that's always an option as a consultant (fixed price).
But as a business owner, I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
If you were digging ditches by hand, but now you have an excavator with you, are you expecting 10x payment just because you can now do it 10x faster? Nobody will pay for that. Instead, you would now charge by a different metric (say, length of the ditch, rather than hours). And you would now be able to handle more customers since you can provide the same outcome to more people in less time.
I have no problem if someone wanted to charge by outcome. I think the concept of paying for "time" is going to go extinct in this new world. Think of lawyers who bill $500/hour. Now they can't say "I need 4 hours to have my team review this" and send you a $4k bill for their whole team to bill, because expectations have changed.
But framing this as an "us vs. them" thing by comparing apples and oranges isn't going to get us anywhere.
> With almost all traditional employment, people are paid for their time. Their output is more an expectation based on the time put in.
Yes and no. It's true that employees are rented for a limited amount of time rather than owned like tools. However, employees can earn vastly different sums of money for the same amount of time. Some make $7 per hour, while others (a privileged few) make $7000 per hour or more.
> I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
Ok, but are you going to let employees who use these "labor-saving" tools work 1/4 of the hours per week as before for the same compensation as before? I suspect not.
> But as a business owner
> But framing this as an "us vs. them" thing by comparing apples and oranges isn't going to get us anywhere.
Of course that's what a business owner would tend to say.
My point is that there's no inherent reason why the benefits of "increased productivity" would trickle down to employees. Individually, employees may look at A.I. and think it's great, but collectively, if all employees have access to the same tools, then employees don't benefit from A.I., because it doesn't fundamentally change their relationship with employers.
Think about how students started using A.I. to cheat and get ahead. This turned out to be short-sighted, because everyone has the same access to ChatGPT, and thus A.I.-fueled cheating became ubiquitous in school, and the cheaters are back where they started, no longer ahead of their peers and no better off than before. Indeed, they may be worse off now, because potential employers have to expect that all college graduates are incompetent cheaters who didn't learn anything.
This is not true at least with the current crop of AI.
I have seen 4-5 people use the tools (won’t say where and in what capacity, but only one of them was an employee of mine) and the results were wildly different.
One person just copy pasted AI responses and added no value. But that meant they also copy pasted substandard responses and it actually reduced the quality of their work (in fact, their greatest asset was client management)
Another one was a coder, and even though he was using windsurf every day, he was getting stuck on the most basic items. And having access to AI made no difference in his productivity because he doesn’t have the base knowledge and experience to be able to discern what is a quality suggestion and what isn’t. And he doesn’t know what questions to even ask the AI to help unblock him.
I actually think these tools will WIDEN the gap between what a good engineer can do, and what a mediocre one can. And it will widen income gaps. I’m not going to say this is a good thing. I’m still a human being.
But I’m just reporting what I’ve observed.
What you said didn't actually contradict me.
For example, with regard to students, it was the mediocre, lazy students who started using ChatGPT to cheat. The good students didn't need it. And as I said, the cheaters are no better off than before. They mistakenly thought they had some completitive advantage, but the same tools were available to everyone.
With regard to engineers, the issue isn't the difference between good engineeres and mediocre engineers. The issue is that all good engineers have access to the same AI. Thus, engineers who were equal to each other before AI are still equal to each other. It's not a competitive advantage over peers, and thus there's no incentive to pay them more. AI doesn't change the supply of good engineers, nor does it give individual good engineers a competitive advantage over other good engineers. Any additional productivity comes more or less for free, because it's due to factors external to the engineer rather than internal to the engineer.
In the end, it's all supply and demand. That's not to say it has no effect either. Increased productivity from AI can increase wages by raising the upper bound for salaries (companies won't pay an employee more than the value he can create), and creating competition between companies (if a single programmer can generate $500k worth of value, and their salaries are only $100k, then we'd expect other companies to spring up to arbitrage this difference).
> In the end, it's all supply and demand.
That's true.
> companies won't pay an employee more than the value he can create
True, but the wealthy corporations aren't simply paying for "programmers". Cheap programming labor is available, especially in less wealthy countries, and corporations could avail themselves of that if they wanted (and they sometimes do). When some programmers make more than their peer average, it's because the employers consider those programmers to be "the best", the smartest, most skilled, hardest working, or whatever. In that regard, A.I. is largely irrelevant. The best programmers and the worst programmers can both use A.I., so it's not a distinguishing factor in the labor market. What matters the most is the supply of the type of programmers that companies are seeking.
A.I. might (or might not) increase a company's revenue, but there's no reason the company can't just book that as profit. A.I. doesn't increase the demand for programmers; if anything, it might reduce the demand for programmers, if the programmers they've already hired become more productive. Adding more employees has diminishing returns, as we've seen with mass tech layoffs in recent years. Hiring is not an automatic productivity boost, if you're measuring productivity as an addition to revenue. At the end of the day, producing more stuff doesn't matter unless the company can sell that stuff.
> if a single programmer can generate $500k worth of value
How do you even determine the monetary value produced by a single programmer in a large team effort?
(Ask yourself this: how much "value" is generated by human resources or administrative assistants or janitors? You might say $0, but nonetheless they get paid more than $0, because the work needs to be done. You can't a functional corporation without them. And their compensation is based on supply and demand, not on "value generated".)
Let me put it this way: I used to work for a small company, and some of my coworkers also worked at a BigCo at some point, either before or after SmallCo. Their personal productivity didn't change much over time, from one employer to another, but their employer's revenue per employee changed dramatically from one employer to another. A tech company's revenue depends on many different factors, many of which have nothing to do with the employed programmers.
Let me put it another way: the more that employers rely on publicly available automatated tools—for example, LLMs—rather than on the individual skills of employees to increase productivity, the more the employees themselves become interchangeable cogs in the machine, and thus by supply and demand, their compensation will not increase.
Of course, if using LLMs were an elite skill that required inborn talent and/or years of specialized training, then you would expect LLMs to increase the compensation of those with that unique skill, but I've seen no evidence that this is the case.
> But as a business owner, I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
s/these tools/drugs
And before you claim that it's an "apples to oranges" comparison, let it be known that we have no data on the long-term cognitive effects of offloading thought to machines.
You’re going to pay almost 4x for an experienced excavator operator vs. a ditch digger.
Not too unlikely the same thing happens with AI.
Maybe because of supply and demand, but not because they bill hourly, which is exactly my point. It won't be simply because the excavator can get it done 4x faster.
I mean that the price multiplier isn't based on the productivity multiplier, as the parent was implying. Of course, the cost of the excavator is built into the price too.
Also for smaller jobs (like with coding), I might now be able to hire a mini excavator (using AI as an analogy for the excavator) myself for a weekend and just do the job myself, and leave the bigger jobs to the pros. That's already happening (look at RR Buildings on YouTube, two guys build massive structures with the help of tools).
> But as a business owner, I would absolutely expect my employees (and vendors) to use these tools to go faster, just like laptops make them go faster than typewriters. It's not like it's causing the employee 4x more effort for that 4x output, is it?
what this means is that the business owners capture all of the benefits of the increased productivity (which is what we've been seeing for decades)
You can keep seeing this through an adversarial lens. And I could just respond with some flippant remark like “well, they’re the ones that pay the bills” - but all that does is create a false battle between parties that have a shared interest (that of making a living and providing value)
Either way, I do care tremendously about the power of efficiency. That can sometimes lead to better profits, but more likely leads to happier customers and more business, which helps the employees.
>what this means is that the business owners capture all of the benefits of the increased productivity (which is what we've been seeing for decades)
Not really. Even though labor's share of GDP has steadily dropped for the past half century, it's on the order of 4% for the past 70 years. Real GDP per capita on the other hand has more than tripled in the same time period.
https://fred.stlouisfed.org/series/LABSHPUSA156NRUG
Because there's an enormous and growing gap between top compensation and lowest quantity pay, a better comparison would be to look at _median pay_ as a share of _GDP per capita_.
"business owners" by definition, means capital. AI engineers, quant traders, and bankers making 7 figures TC doesn't magically mean they're "business owners" in "business owners capture all of the benefits of the increased productivity".
The tools have gotten a lot better for sure. Still not good at scale, but for narrow things that are well solved, they crank it out.
>> still wonder how well it would work against a large established repository where it isn't free to re-invent the wheel and doesn't know about established architecture and methods etc..
You’re missing the point of AI coding if you think that’s the measure of its capabilities.
I admit, have a negative feeling that if my Computer Science discipline is going down in the Titanic, then other disciplines should go down with me. I know its not a good attitude. It feels unfair that CS job will be the first to fall, but we have had it so good for so long, maybe it is fair. Our industry moves so fast. I suspect lawyers, doctors, other "white collar" jobs will take longer and be slower to disrupt due to societal delays and protections (unions, laws, etc.), even though AI could potentially affect these disciplines more down the road.
The number of jobs doing low-level code will likely increase even as AI reduces the premium for the skills.
Eh most of the jobs that will survive over time are already shit jobs. I cant wish too ill on people in those situations.
https://archive.is/msi41
So far AI seems to better in generating technical debt than value. Not exactly worried - in 3-4 years there will be a lot of investment in clearing up the currently created vibe mess.
Do we get a mass layoff of "nocode"-mbas? That would be a appropriate response to "i tanked the company to remove a dependency ".
So far that's what most of the layoffs have been.
“Vibe cleaner”
AI is, currently, coming not for the coders who made it but for the coders who didn't contribute to or ignored it. The foundation labs are all quite committed to recursive self-improvement of coding tools as a general research accelerant.
When I see how unsupervised LLMs struggle with coding tasks (see public PRs from Copilot on Microsoft codebases) I don't see how recursive self-improvement can lead to any actual improvement rather than the total opposite.
Far from the truth.
If anything, more than ever you need to be a great programmer to know how to direct the process to a worthwhile outcome.