Replacing Contractors with AI
Posted by Firebefore40@reddit | ExperiencedDevs | View on Reddit | 55 comments
I am not sure how common this is, but one of our engineering manager talks about replacing contractors with AI agents. Their idea of ideal team looks like bunch of staff engineers with AI agents around them instead of contractors. This got me thinking couple of questions.
-
Is this the end of an era where outsource comes to an end.
-
Why staff engineers ? What about senior engineers ? They are like slightly better agents ? If so what happens to industry in 10 years where these staff engineers retire ? Because it looks like their expectation is that junior roles can be filled with agents. Is this really a good sustainable practice for a company.
Less-Ad5766@reddit
This guy is an idiot.
Do agents even can do anything meaningful without swe's design and reviews?
thephotoman@reddit
I’m unconvinced they’re actually making me more efficient, just that they’re turning the job of writing code into a job of reviewing their outputs and telling them that they are wrong.
Main-Drag-4975@reddit
I ship more and lower quality stuff now. Sadly I work at a place that is happy with that, not that I think they fully understand what they’ve lost.
yubario@reddit
You’re intentionally choosing to ship lower quality content. AI is smart enough to follow refactoring plans and test cases. There is no excuse to poor code quality other than laziness, just like it was before AI.
psaux_grep@reddit
«AI is smart…»
No, it’s simulating smartness within the framework of a probabilistic engine.
AI is definitely a benchmark for a hire (the person would need to perform better than AI does in terms of thinking), but let’s not for one second imagine that AI as it’s currently implemented/delivered is actually thinking.
nextnode@reddit
It is reasoning by the definition of the term and its use in the relevant field including relevant research.
I would also argue it is already smarter than most redditors.
FetaMight@reddit
What does Turing completeness have to do with anything here? Are you thinking of the Turing Test?
And, last I checked, AIs based on LLMs, even in agent swarms, still didn't meet the criteria for a "reasoning system" by the definitions of "reasoning" in psychology and its related definition in computer science.
If you think "some level of reasoning is nothing special and decades old." I would love to see an example and/or scientific research backing that up.
SJrX@reddit
While the original post mentioned "meaningfully" lower quality stuff. I dunno, while I mostly do stuff with Claude right now, and haven't coded meaningfully in 6 months. I can't help but feel that that while I might be churning out 10x the code, the quality is less. I personally think the value trade off is there, 10x more code at 90% the quality is a good trade off in my view.
How are you making your code, as high quality as before. There is a big difference mentally for me from being in a flow state churning through code, and just carefully review code with less context. I honestly would be a bottle neck to carefully reviewing all code, and am currently trying to find ways to make the code more reliable (e.g., SpecKit, Better more explicit tests), etc... I don't think I have an answer at this moment.
yubario@reddit
Again, it is your choice to decide sacrificing quality over quantity. You are making the executive decision to accept code that has less quality because you’re okay with making that decision.
This is no different than how it was before AI.
Literally nothing has changed.
If you’re telling me that the engineer has no impact to the quality of code generated then that means cheap offshore labor can just replace us overnight. Except they can’t, because the engineer does matter.
SJrX@reddit
Right but what I was responding to is there being "no excuse to poor code quality", which I think you are saying yes velocity is a trade off you can decide, just like before AI.
AI is meaningfully different also because it changes the economics and calculus of the decision, and we don't yet have experience in exactly how the quality suffers.
If I have manufacture "high quality" lightbulbs and can do 10,000 a day with 100 fails (1%). If expand or change how things work, and now can do 100,000 a day but have 5,000 failures a day, that's a 5% failure rate, and maybe no longer high quality. But depending on how much failures costs still way more profitable. You are forced by economics down this road, and it might bother you if you were the high quality company before.
I'm not saying that at all, I'm not sure we are getting replaced ... yet. Right now I think the limiting factor is that LLMs are limited by the amount of context we can give it. Even if the LLM made less unforced errors on it's side, I don't know if we are going to be able to give it enough uncontradictory things to completely get rid of the dev. That said, I don't know how we are going to stay engaged in the process, writing code and actively engaging in it, means we know it and think about it much more deeply than passive reviews.
BigBootyWholes@reddit
This sub doesn’t realize AI is the king at automating things. Like I’ve developed a pretty advanced workflow that automates 99% of bug investigations in your production code. People here tend to be very “book smart” but aren’t very good at developing unique solutions or thinking outside of the box.
Constant-Tea3148@reddit
If the company prioritizes velocity over quality because someone convinced leadership every engineer is now "10x" then I think there is an excuse, they're just fulfilling unreasonable expectations to the best of their ability.
yubario@reddit
I get what you’re saying but you can always run more than one agent at a time. As you’re vibing one project, while waiting, vibe another project and make it cleaner.
There is always time for cleanup.
Even if you ship code untested there is usually some form of downtime during your day where it’s possible to do this. If not, find another job because they’re overworking you.
ItSeemedSoEasy@reddit
You sounds like you're on a train straight to a town called burnout.
yubario@reddit
Running two agents doesn’t cause burnout.
I have to read the code either way and fix it, regardless if it’s an AI agent doing it or not. I’m still reading the same amount of code I’ve always done.
xXxdethl0rdxXx@reddit
That's oversimplifying the issue. What we're seeing is a reluctance to add or replace human headcount by virtue of velocity being anticipated as higher—meaning that the expectations of individual output are higher.
When that happens, you are telling your engineering teams that speed is more important than quality. This isn't new or unique to AI in any way.
BigBootyWholes@reddit
Do you guys really outsource work and not review it anyways?
I swear this sub is full of bad devs.
Inevitable_Zebra_0@reddit
Most devs are low-quality or procrastination-prone. There're usually 1-2 gems per team, holding the whole project on their shoulders, while the rest are mediocre at best. At least that's the pattern I've been observing throughout my career, long before AI.
thephotoman@reddit
The industry is full of bad devs. I have some coworkers in a nightmare scenario because of a lot of sloppy coding practices that predate AI.
I have a patchset to bail them out, because I’d seen this particular sloppiness before, and I knew how to fix it. It’s still stuck in their backlog, though.
tevs__@reddit
I mean, at the top end of IC, that's predominantly the job - creating tasks for others and reviewing their output to make sure they understood and delivered the brief.
ItSeemedSoEasy@reddit
There's a reason it's the top end. Because only the top end people can do it well.
johnpeters42@reddit
Look up "reverse centaur"
dbxp@reddit
Have you seen the code quality of the big Indian outsourcing firms?
ham_plane@reddit
Replace "agents" with "contractors" and it's the same sentence
(I'm assuming we're talking about transient/overseas, non-expert contractors)
Firebefore40@reddit (OP)
I think this is where the idea of having staff engineering manning agents comes from
CraZy_TiGreX@reddit
This does work, it ends up burning out your best engineers, but in the meantime it works.
And eh, that's kpis, profits, etc in the meantime, as a manager you might get a promotion and all!
About 10 years, nobody knows the future, I don't think juniors will be replaced by agents on their own, even that, you will need seniors but no juniors leads to no seniors, etc. Anyway, not my company.
brocodini@reddit
Kick your engineering manager out of the building.
Firebefore40@reddit (OP)
Well most likely he would get a promotion for his genius idea from top management.
octatone@reddit
Well, until top management gets the usage bill. LOL.
johnpeters42@reddit
This guy BOFH's
w-lfpup@reddit
Your engineering manager is a libertarian POS that would pay someone in rocks if it meant saving a few bucks that's why he's hiring contractors in the first place.
AI is just a tool of the bourgeoisie to cut everyone's pay in half, a "look at this hand" while the stiff you with the other
Idea-Aggressive@reddit
Wouldn’t a contractor be a libertarian too?
w-lfpup@reddit
As a former contractor I can tell you with 100% certainty that every contractor would rather be a full-time employee with benefits and health insurance, they'd even take a government job
Sworn@reddit
Highly dependent on the country. In my country self-employed contractors have very beneficial tax codes, so you generally end up making way more as a contractor than an employee.
Idea-Aggressive@reddit
That’s basically in most countries. The problem today is that there’s not much work and interview process for contractors is as lengthy as permanent; quite unfair given how short some contracts are, plus having to bother people with references which is humiliating
w-lfpup@reddit
Cool!
Idea-Aggressive@reddit
The context was the libertarian statement.
I’ve mostly worked as a contractor for the past two decades. With contract lengths as short as 10 days, and having to go through same interview process as everyone else’s: 5, 6 stages etc.
Just trying to make a living.
fsk@reddit
Nowadays, if you are hiring an offshore team, odds are they are doing everything with AI. They probably aren't even reading and writing their E-Mails, using AI for that also. You might as well cut out the middleman and just use AI.
Most places are doing the opposite. Instead of firing their cheap workers and replacing them with AI, they're firing their most experienced/expensive workers and hoping that a cheap offshore team with AI is just as good.
PTTCollin@reddit
I'd take agents over offshore contractors any day of the week. You need highly skilled engineers at the helm, but that's not what they're talking about replacing.
Fit-Notice-1248@reddit
In my experience, offshore contractors will take the simplest requirement and it extrapolate into the most spaghetti code you could ever witness.
hibikir_40k@reddit
It all depends on how much you are actually asking of the contractors, and how much they really own their work. I sure have worked at places where the median contractor was worse than posting tickets to an LLM, as they didn't actually attempt to own their work in the slightest. Deploy to prod? Basic testing? Nope, they sucked. I've seen people spend two weeks in what took Claude 15 minutes. If that's the standard you are getting, sure, replace away.
The thing is that if you are hiring that cheap, and your culture is that bad, you should change a lot of things anyway, but one can have good contractors that you can delegate high level things, and then the AI agents don't win at all.
dbxp@reddit
Imo this is an interesting proposal for AI. It's certainly not perfect but when you compare to the results you get from outsourcing to Wipro or Tata I think it's pretty reasonable.
I'm guessing they used staff engineer to just mean anyone experienced.
If_I_Could_Just@reddit
Not sure I’d mind this. Working with AI agents does feel somewhat like working with a contractor who won’t do much more than asked. Except I don’t have to care about hurting the agents’ feelings when it has to redo something.
Resident-Trouble-574@reddit
Maybe one day AI will take over and you'll realize that you should have had to care about its feelings.
sarhoshamiral@reddit
Considering right now "memory" of an agent is a text file I am not worried. Now if we train next set of models with those memory files then continue to that loop, we may actually see models personalizing output based on small hints in your prompt, context. Ie model actually remembers you.
alephaleph@reddit
They thing I see coming is outsourced contractors who employ AI. Cheap babysitting for the AI that leadership loves with all the velocity gains… until the total lack of context expertise the project causes everything to go boom.
TrainingDragonfruit1@reddit
What is stopping you from replacing all engineers with AI? Stopping at contractors seems like conservative move. /s
NeuralHijacker@reddit
Low tier overseas contractors? Yes absolutely. AI is far better.
casualPlayerThink@reddit
Agents and generative AI should be given for workers as a tool, not replace them. It will be more expensive, and the slop is guaranteed. Also, the PM should learn to phrase and specify everything that they rarely do...
No
In some cases, current agents can produce as good results as a junior/intern would. But it can not replace creativity, critical thinking, and intelligence. Staff engineers coz' they manage agents like a team or bunch of juniors and other coworkers (I think).
funbike@reddit
We are not yet in the trough of disillusionment of the hype cycle for AI codegen tools. It's coming, though.
But to address OP's post, as with any big change, it should be piloted at micro scale and then scaled up gradually, based on comparative metrics and milestones. Anytime I hear anybody talking about anything big (not just process/coding/AI) being changed quickly, I know it's going to be a failure or extremely painful. Change should always be done slowly.
I think juniors are going to have a hard time getting jobs over the next 3 years or so. But I think that will reverse. The industry hasn't yet figured out what skills juniors need to prioritize in an AI workplace to get a job and survive long enough to become a senior.
mq2thez@reddit
In a typical large software company, staff engineers are those who have started to get to the point where they’re more effective making large scale changes, planning work for multiple teams, thinking ahead many months, etc. People who are starting to hit the point where they’re having more impact by guiding teams of people rather than directly going hands on themselves. It’s even more pronounced at Senior Staff.
So AI-pilled leadership think they can cut out a lot of the “fat” by having those staff engineers guiding teams of AI bots rather than teams of people.
It’s fucking dumb, because not all staff engineers are actually built that way. Additionally, the whole model of being able to plan/strategize and hand work off depends heavily on the people you’re handing work off to being actual people with actual brains and experience. If I have to fully guide the process like with AI, I’m not actually being nearly as effective as I could be. AI is light years away from being able to be a true handoff, so it’s less like I give work to a team and more like I have to sit in a room where a bunch of junior engineers constantly need me to pair program with them.
bold_snowflake@reddit
Does this engineering manager write any code themselves?
While I agree with a previous poster that this guy is an idiot, I think there is more too it than that.
In my company I'm an engineer manager running a few teams, but I do actively code still on a daily basis. One step of management above me doesn't write code anymore. Anecdotally I'm seeing the people that don't write code are making decisions detached from reality.
mpanase@reddit
Honestly, stupid idea.
Sad to hear it comes from an engineering manager, who should have certain knowledge about the topic.
You can't stop it, though.
They will only learn thorugh pain.
Encourage them. But remember that it's not your job to make save them from their unstoppable stupid ideas; let them crash and burn.
disposepriority@reddit
Talk is cheap - do it now and come back in a year to let us know how it went!
Bricktop72@reddit
We're comparing MSP work to AI work. We've replaced a few offshore teams with a senior+AI with success. We just haven't tried it at scale.