It’s wild to me how many software engs never want to see code again. IMO they should just retire so someone else that does want to look at code can take the job. The laziness is unreal, all while the same people acting like they’re productive.
I get paid to solve problems, not to write code. Before writing code was required to solve the problem, it's not a hard rule anymore.
Like a carpenter won't get more money for their cabinets if they make them all painstakingly by hand, not using a single power tool. (OK, some do but it's more of an art piece than a functional thing).
Carpenters can and do make more money when they make things by hand. But the customers base for this is indeed smaller.
It's like complaining about a chef who make all their food from fresh ingredients instead of buying off-the-shelf processed versions. People will pay more for that freshy made food.
This is my problem with it, I can’t tell if people actually think the LLMs writing all their code is the future or they just want it to be the future because they’d prefer not to think and still get a dopamine hit from it.
I got into this field because it's the highest paid job I have aptitude for, not because I was ever particularly excited to write jobslop code that consists mostly of basic CRUD and business logic.
I don't act as a meat proxy since I have some respect for myself and my coworkers. However, whatever interest in programming I ever had, was completely gone after a couple of years on the job.
That being said, I am, in fact, speedrunning early retirement.
Thanks for your honesty. It explained some things I watched before retiring. I am one of those people who loved programming and hated what corporate settings made out of it. Yeah, it's one of the better paid things. It's hard work, and there are things that pull it in exactly opposite directions. Sometimes even for simple things like CRUD, when something is requested that would go sideways, you have the choice of rejecting it and avoiding thinking hard, or taking it by the horns. I told my wife, sometimes you just can't leave, and it's not only emergencies. Because if you are in a deep session and you stop, you lose your own context. But it's very satisfying if you can wrap it up in a way that leaving notes for yourself for later suffices. But this style of work is not for everyone. And in corporate settings weird things happen, and sometimes I just felt shat on. But hell, it's paid well, right?
Okay, now that that is out, let me ask politely: in what way do you avoid acting as a meat proxy for AI?
EDIT: re reject, you can do that while acting as accepting it, by saying something like, yeah, let's do that but before starting, we will look at the problem, it's a game of shifting responsiblity to other people. For me that was a corporate game that has nothing to do with programming.
The linked post describes being a meat proxy as copypasting tickets into AI and pushing the output without reading.
I don't copypaste tickets to AI, I decide on how each task is solved and LLMs only implement what I decided on. I read the code before making a PR. I review other's PR with my own eyes. I never copypaste LLM output to my coworkers in chat. What else is there to it?
It's fine. I still program for fun. I switched to Pi as harness, and instructed the model to careful stop early such that I can stare at code, and I tend to give extremly substantive feedback sometimes. It was not easy, and now I hope I found something that clicked for me. I modified the bash tool to force the agent to tell me what he does, and it helps wonders, the agent says, he wants to do X and I can verify if the bash commands do what the agent says. I also use lightweight sandboxing. I am happy that no financial pressures influence my use of AI. What's still open for me is how to discuss it publicly, since there's a large group of programmers jumping at programmers who admit they use AI.
Maybe it's just the sign of an aging profession. A combination of biological aging and an accumulated cynicism sets in, and the expert will declare their field dead and organize its funeral. For a physicist that might hold unwarranted skepticism for newer theories, and a software engineer might attempt to kill their field all-together with AI, visual coding and what not, pulling up the ladder behind them.
What if I, as someone employing you, don't specifically need someone to look at the code?
We may not be quite there yet, but what if automations using LLMs can do it faster than software engineers who look at the code can, and to an acceptable standard?
> all while the same people acting like they’re productive.
Are they delivering features, capabilities, updates, refactors etc? Faster than those who are "looking at the code" ?
Don't get me wrong, I've spent 25 years reading and writing code, I'm pretty sure I'm quite good at it, though I wouldn't claim to be any sort of superstar. My job has changed immeasurably in the last year, and I'm starting to wonder if "reading the code" won't go the same way as "verifying the assembly that the C compiler spits out to make sure it's correct"
A compiler is deterministic, even if you don’t understand the assembly at least it will do exactly the same thing every time. I absolutely loath this comparison.
I haven't looked at the author's age, etc. But calling for someone to be fired is "not okay". I understand that we live in a polarized world and I understand the frustration but it's still "not ok" for this titles to float around. They feed the beast and that's that.
ps. The assumption is that there was empathy at play to begin with, which realistically speaking many people don't have. They would be calling for co-workers to be fired on several grounds: merit (or "merit"), skill, color-skin, dna, etc.
> Wasting your time... Forcing you to do their work for them
The one forced to do their work for them, are the teammates. Not the manager. The manager only care of it to be done. How it is done: doesn't matter.
I would argue that one of the main reason for meat proxy, is security boundaries.
In the majority of companies with a certain scale, no one can run an AI agent that will do the project from A to Z. You always need to raise at one point ticket to someone else, a human, it cannot even be done by an automated system (technically it could of course, but acceptable by governance... unfortunately no) .
This is written for devs but the same applies to many areas of knowledge work. In my company one of the negative side effects of widespread AI adoption is that the worst performers or people with the least understanding of what’s being done (and typically generally lazy if we’re being honest) have adopted AI tools to do everything- review and summarize documents they should be reading, drafting emails and design documents, and in many cases writing messages to team members.
The company seemingly has no issues with this as it’s almost always blatantly obvious that AI was used for the output.
For example, Copilot has a very distinctive and obvious color scheme and format it uses for Word documents and diagrams that are almost always orders of magnitude more complicated than is warranted (which makes all of this coming from people who have zero understanding of what we’re doing even more absurd).
Nevertheless they do it every day and send AI-generated summaries of documents out to the team for review without even reading them first. I’ve railed against this for nearly two years to the point it’s nearly driven me mad but so strong is the AI adoption religion that suggesting that it’s poor practice feels like blasphemy and by suggesting real work be done that I’m somehow doing things the wrong way.
This has allowed these types of people to remain in their jobs and thrive and so when the AI bubble finally bursts I fear it will reveal largely hollowed out companies where top lower level people have either been pushed out or left.
This exactly matches my experience over the last two years. Leadership and investors seem to love the good vibes from the performative shipping of slop, too excited by all of the movement to notice the lack of actual progress.
That this is still happening seemingly without organization-level consequences has me questioning my sanity. Does anything we do matter enough that doing it badly with lazily prompted LLMs causes problems? I was sure the answer was “yes” but looking at the bottom-line investor perspective it’s hard now to make that argument.
Now that long running agents like Dots are available without any effort to set up I think the days of interacting with actual humans over Slack or any async medium are soon to be in the past. This can’t end well…
In my experience, this kind of developer was always the one taking shortcuts, generating bad code and tech debt and in general just thinking about the ticket they were working on and nothing else.
This post completely misses the point and experience that you get as a person who works for a company. There is always an over exaggerated expectation of delivery. You're ranked on how much you can deliver and it completely ignores what is humanly and reasonably possible.
LLMs and agents allow for you to deliver at the rate. But you also have risk associated with that. If you don't like the output, fix your culture before you start blaming people for being "meat proxies".
I have worked for a few companies in my life and that just wasn't my experience in any of them. People tend to trust your judgment when you show that you know what you're talking about, and the use of agents to generate your output seems detrimental to that.
I check the results and test the agent's changes but for the most part my job now is being a meat proxy. All of the things I get to work on were completely trivial for me to implement so they're even more trivial for the current agents. I don't really see how I can add value to it, like some things are just "hook up this button to call an API", even the models a year ago couldn't fuck this up. At this point the only thing I'm really writing is PR descriptions.
I wouldn't mind being fired at this point though, my job feels like a joke.
Meh. Management encourages it. Just do what management wants and collect your paycheck. In this age the nail that sticks out gets hammered down, don't burn yourself out trying to rise above it. And if you think it's bad now, it's only going to be worse in a decade when the next generation come along who have only ever done work with AI.
Then sending an agent’s result can send a message. And being blessed with great a great team I may get feedback on what caused this to be skipped or fail. We don’t spam each other me forwarding is either endorsing it or asking a question on validity of findings for a reason.
But in any case, we are all heading there? Eventually, a SWE will be "a nice smart person that's more pleasurable for the manager to speak to than LLM itself".
Imho, 90% of "tech" jobs were mindless bs. AI is just exposing it at greater efficiency. The future will be, sadly, mass layoffs and greater corporate profitability. It will drive demand for kind, very smart systems level thinkers that can understand how to mass scale LLMs to build as though they are teams of 100 engineers. I don't know where this ends, but I think we will be living in a very different world in a few years.
As someone who made a reasonable reputation for himself by being a meat proxy for google search for years (Hey, go ask him, he knows about loads of stuff...) I'm surprised anyone even notices if others are just doing auto-reviews.
> I always saw the software engineering field as being comprised of two groups of people: people who are devs because they are passionate about building software, and people who don't actually enjoy software but joined the field because it's a job that pays well.
There's a third motivation hiding in there, people who were passionate about coding, rather than building software. It's a subtle difference but an important one. Folks whose primary focus is building and delivery are having a field day with LLMs because they've found a way to build and deliver in far less time with far less resource. Folks who have been more passionate about the coding itself are finding that their legs have been shot out from under them, because what they enjoyed and where they thought their value lay is becoming less relevant.
I absolutely get the frustration here, but conversely I think some companies now are seeking to remove the human from the loop as much as possible. That interactive AI-assisted workflow? That's a mid-state, why not instead have a multi-stage automation pipeline, to digest a set of requirements, build granular jira tickets for the work, come up with acceptance criteria, implement test harness additions for those, build the code, test it, iteratively review the code from several different viewpoints (security, minimalism, performance etc), and post it as a PR? Sure, you probably want a plan-review checkpoint, and the agents should stop and ask when they need to know something from their human overseers, but ... we're going more automated in general.
Morally speaking, the AI was trained on our code. It's our right to be meat proxies, so long as we deliver.
Also, practically speaking, if the codebase is well designed and has good foundations, being a meat proxy might actually work. If someone puts careful, meticulous thought into getting the foundations of the project right and then turns into a meat proxy because the thing is so well laid out, it practically builds itself, should they be fired?
If this is the case, then developers have the incentive to create a rotten overengineered foundation, then the AI won't be able to get anything done without human guidance. Great way to make yourself indispensable as an engineer.
What the author is suggesting isn't only immoral, in practice, it's not going to work. It's why we're in the mess we are in now with projects with millions of lines of code with new critical vulnerabilities popping up daily. Show me the incentive, I show you the outcome.
Managers who promoted people based on how much code they wrote and how many PRs they merged are going to get their punishment soon. Whatever metric you use to coerce engineers into working in a particular way is going to create problems later. You can't coerce people into producing quality engineering.
If you ask an experienced farmer to plant a crop but you insist that they must dig holes using their nose because that's your way of doing things, don't expect them to be productive and don't expect them to stick around very long.
I do this sometimes, but with things that are anyway useless crap nobody cares about. Or even worse where the task is small, but someone used Claude to write a huge ticket. I literally cannot bother reading that slop.
It’s wild to me how many software engs never want to see code again. IMO they should just retire so someone else that does want to look at code can take the job. The laziness is unreal, all while the same people acting like they’re productive.
I get paid to solve problems, not to write code. Before writing code was required to solve the problem, it's not a hard rule anymore.
Like a carpenter won't get more money for their cabinets if they make them all painstakingly by hand, not using a single power tool. (OK, some do but it's more of an art piece than a functional thing).
Carpenters can and do make more money when they make things by hand. But the customers base for this is indeed smaller.
It's like complaining about a chef who make all their food from fresh ingredients instead of buying off-the-shelf processed versions. People will pay more for that freshy made food.
But to call one or the other 'lazy'? It is just different.
Nature is lazy. And as long as they take responsibility for their merge requests (and understand them) ..
This is my problem with it, I can’t tell if people actually think the LLMs writing all their code is the future or they just want it to be the future because they’d prefer not to think and still get a dopamine hit from it.
And how would they understand them if they didn't even read them?
Cannot agree more.
I got into this field because it's the highest paid job I have aptitude for, not because I was ever particularly excited to write jobslop code that consists mostly of basic CRUD and business logic.
I don't act as a meat proxy since I have some respect for myself and my coworkers. However, whatever interest in programming I ever had, was completely gone after a couple of years on the job.
That being said, I am, in fact, speedrunning early retirement.
Thanks for your honesty. It explained some things I watched before retiring. I am one of those people who loved programming and hated what corporate settings made out of it. Yeah, it's one of the better paid things. It's hard work, and there are things that pull it in exactly opposite directions. Sometimes even for simple things like CRUD, when something is requested that would go sideways, you have the choice of rejecting it and avoiding thinking hard, or taking it by the horns. I told my wife, sometimes you just can't leave, and it's not only emergencies. Because if you are in a deep session and you stop, you lose your own context. But it's very satisfying if you can wrap it up in a way that leaving notes for yourself for later suffices. But this style of work is not for everyone. And in corporate settings weird things happen, and sometimes I just felt shat on. But hell, it's paid well, right?
Okay, now that that is out, let me ask politely: in what way do you avoid acting as a meat proxy for AI?
EDIT: re reject, you can do that while acting as accepting it, by saying something like, yeah, let's do that but before starting, we will look at the problem, it's a game of shifting responsiblity to other people. For me that was a corporate game that has nothing to do with programming.
The linked post describes being a meat proxy as copypasting tickets into AI and pushing the output without reading. I don't copypaste tickets to AI, I decide on how each task is solved and LLMs only implement what I decided on. I read the code before making a PR. I review other's PR with my own eyes. I never copypaste LLM output to my coworkers in chat. What else is there to it?
It's fine. I still program for fun. I switched to Pi as harness, and instructed the model to careful stop early such that I can stare at code, and I tend to give extremly substantive feedback sometimes. It was not easy, and now I hope I found something that clicked for me. I modified the bash tool to force the agent to tell me what he does, and it helps wonders, the agent says, he wants to do X and I can verify if the bash commands do what the agent says. I also use lightweight sandboxing. I am happy that no financial pressures influence my use of AI. What's still open for me is how to discuss it publicly, since there's a large group of programmers jumping at programmers who admit they use AI.
Maybe it's just the sign of an aging profession. A combination of biological aging and an accumulated cynicism sets in, and the expert will declare their field dead and organize its funeral. For a physicist that might hold unwarranted skepticism for newer theories, and a software engineer might attempt to kill their field all-together with AI, visual coding and what not, pulling up the ladder behind them.
,s/code/machine &/g
Your job was never code.
What if I, as someone employing you, don't specifically need someone to look at the code?
We may not be quite there yet, but what if automations using LLMs can do it faster than software engineers who look at the code can, and to an acceptable standard?
> all while the same people acting like they’re productive.
Are they delivering features, capabilities, updates, refactors etc? Faster than those who are "looking at the code" ?
Don't get me wrong, I've spent 25 years reading and writing code, I'm pretty sure I'm quite good at it, though I wouldn't claim to be any sort of superstar. My job has changed immeasurably in the last year, and I'm starting to wonder if "reading the code" won't go the same way as "verifying the assembly that the C compiler spits out to make sure it's correct"
A compiler is deterministic, even if you don’t understand the assembly at least it will do exactly the same thing every time. I absolutely loath this comparison.
Do you think you're better than them? Is it possible they understand something you don't? What does "software engineer" even mean?
I haven't looked at the author's age, etc. But calling for someone to be fired is "not okay". I understand that we live in a polarized world and I understand the frustration but it's still "not ok" for this titles to float around. They feed the beast and that's that.
ps. The assumption is that there was empathy at play to begin with, which realistically speaking many people don't have. They would be calling for co-workers to be fired on several grounds: merit (or "merit"), skill, color-skin, dna, etc.
> Wasting your time... Forcing you to do their work for them
The one forced to do their work for them, are the teammates. Not the manager. The manager only care of it to be done. How it is done: doesn't matter.
I would argue that one of the main reason for meat proxy, is security boundaries.
In the majority of companies with a certain scale, no one can run an AI agent that will do the project from A to Z. You always need to raise at one point ticket to someone else, a human, it cannot even be done by an automated system (technically it could of course, but acceptable by governance... unfortunately no) .
This is written for devs but the same applies to many areas of knowledge work. In my company one of the negative side effects of widespread AI adoption is that the worst performers or people with the least understanding of what’s being done (and typically generally lazy if we’re being honest) have adopted AI tools to do everything- review and summarize documents they should be reading, drafting emails and design documents, and in many cases writing messages to team members.
The company seemingly has no issues with this as it’s almost always blatantly obvious that AI was used for the output.
For example, Copilot has a very distinctive and obvious color scheme and format it uses for Word documents and diagrams that are almost always orders of magnitude more complicated than is warranted (which makes all of this coming from people who have zero understanding of what we’re doing even more absurd).
Nevertheless they do it every day and send AI-generated summaries of documents out to the team for review without even reading them first. I’ve railed against this for nearly two years to the point it’s nearly driven me mad but so strong is the AI adoption religion that suggesting that it’s poor practice feels like blasphemy and by suggesting real work be done that I’m somehow doing things the wrong way.
This has allowed these types of people to remain in their jobs and thrive and so when the AI bubble finally bursts I fear it will reveal largely hollowed out companies where top lower level people have either been pushed out or left.
This exactly matches my experience over the last two years. Leadership and investors seem to love the good vibes from the performative shipping of slop, too excited by all of the movement to notice the lack of actual progress.
That this is still happening seemingly without organization-level consequences has me questioning my sanity. Does anything we do matter enough that doing it badly with lazily prompted LLMs causes problems? I was sure the answer was “yes” but looking at the bottom-line investor perspective it’s hard now to make that argument.
Now that long running agents like Dots are available without any effort to set up I think the days of interacting with actual humans over Slack or any async medium are soon to be in the past. This can’t end well…
In my experience, this kind of developer was always the one taking shortcuts, generating bad code and tech debt and in general just thinking about the ticket they were working on and nothing else.
This post completely misses the point and experience that you get as a person who works for a company. There is always an over exaggerated expectation of delivery. You're ranked on how much you can deliver and it completely ignores what is humanly and reasonably possible.
LLMs and agents allow for you to deliver at the rate. But you also have risk associated with that. If you don't like the output, fix your culture before you start blaming people for being "meat proxies".
I have worked for a few companies in my life and that just wasn't my experience in any of them. People tend to trust your judgment when you show that you know what you're talking about, and the use of agents to generate your output seems detrimental to that.
I check the results and test the agent's changes but for the most part my job now is being a meat proxy. All of the things I get to work on were completely trivial for me to implement so they're even more trivial for the current agents. I don't really see how I can add value to it, like some things are just "hook up this button to call an API", even the models a year ago couldn't fuck this up. At this point the only thing I'm really writing is PR descriptions.
I wouldn't mind being fired at this point though, my job feels like a joke.
Meh. Management encourages it. Just do what management wants and collect your paycheck. In this age the nail that sticks out gets hammered down, don't burn yourself out trying to rise above it. And if you think it's bad now, it's only going to be worse in a decade when the next generation come along who have only ever done work with AI.
> literally all they do is prompt the agent "do X" and copy-paste the results back to you
But there are results. Why didn't you do that yourself first then? Why are you wasting meat proxy's time?
How often could the question have been answered by that person’s own agent in the first place?…
Then sending an agent’s result can send a message. And being blessed with great a great team I may get feedback on what caused this to be skipped or fail. We don’t spam each other me forwarding is either endorsing it or asking a question on validity of findings for a reason.
But in any case, we are all heading there? Eventually, a SWE will be "a nice smart person that's more pleasurable for the manager to speak to than LLM itself".
Finally being a nice, pleasant person is paying off.
Humans will always need other humans they can blame. Sending messages to a chatbot in capital letters doesn't give the same dopamine hit.
Imho, 90% of "tech" jobs were mindless bs. AI is just exposing it at greater efficiency. The future will be, sadly, mass layoffs and greater corporate profitability. It will drive demand for kind, very smart systems level thinkers that can understand how to mass scale LLMs to build as though they are teams of 100 engineers. I don't know where this ends, but I think we will be living in a very different world in a few years.
As someone who made a reasonable reputation for himself by being a meat proxy for google search for years (Hey, go ask him, he knows about loads of stuff...) I'm surprised anyone even notices if others are just doing auto-reviews.
> I always saw the software engineering field as being comprised of two groups of people: people who are devs because they are passionate about building software, and people who don't actually enjoy software but joined the field because it's a job that pays well.
There's a third motivation hiding in there, people who were passionate about coding, rather than building software. It's a subtle difference but an important one. Folks whose primary focus is building and delivery are having a field day with LLMs because they've found a way to build and deliver in far less time with far less resource. Folks who have been more passionate about the coding itself are finding that their legs have been shot out from under them, because what they enjoyed and where they thought their value lay is becoming less relevant.
I absolutely get the frustration here, but conversely I think some companies now are seeking to remove the human from the loop as much as possible. That interactive AI-assisted workflow? That's a mid-state, why not instead have a multi-stage automation pipeline, to digest a set of requirements, build granular jira tickets for the work, come up with acceptance criteria, implement test harness additions for those, build the code, test it, iteratively review the code from several different viewpoints (security, minimalism, performance etc), and post it as a PR? Sure, you probably want a plan-review checkpoint, and the agents should stop and ask when they need to know something from their human overseers, but ... we're going more automated in general.
All work will be like this forever more. Most work is actually pointless bullshit.
Get used to, or opt out of it all.
Morally speaking, the AI was trained on our code. It's our right to be meat proxies, so long as we deliver.
Also, practically speaking, if the codebase is well designed and has good foundations, being a meat proxy might actually work. If someone puts careful, meticulous thought into getting the foundations of the project right and then turns into a meat proxy because the thing is so well laid out, it practically builds itself, should they be fired?
If this is the case, then developers have the incentive to create a rotten overengineered foundation, then the AI won't be able to get anything done without human guidance. Great way to make yourself indispensable as an engineer.
What the author is suggesting isn't only immoral, in practice, it's not going to work. It's why we're in the mess we are in now with projects with millions of lines of code with new critical vulnerabilities popping up daily. Show me the incentive, I show you the outcome.
Managers who promoted people based on how much code they wrote and how many PRs they merged are going to get their punishment soon. Whatever metric you use to coerce engineers into working in a particular way is going to create problems later. You can't coerce people into producing quality engineering.
If you ask an experienced farmer to plant a crop but you insist that they must dig holes using their nose because that's your way of doing things, don't expect them to be productive and don't expect them to stick around very long.
I do this sometimes, but with things that are anyway useless crap nobody cares about. Or even worse where the task is small, but someone used Claude to write a huge ticket. I literally cannot bother reading that slop.