Do the thing that AI can't do: Sit down with your clients, talk to them, figure out an exact spec, and then maybe do the visual and UX design by hand? AI's not very good at that currently, compared to a skilled professional. Customer service and design is what makes you stand out these days.
Its exactly the same elsewhere, Ive become the goto for multiple teams because these people would rather ask me questions even when the Claude that our company pays for would answer it better most of the time.
People just dont wanna talk to bots, no matter how good the bot is, when theyre talking to a bot and not a human they feel ignored.
There will be lots of work for people who can fix vibecoded messes for some time. You could lean into that market.
If you're a coder, learn to use AI effectively, meaning: The end result is equivalent quality to a hand written codebase. It's far slower than vibecoding, but still faster than it used to be. Maybe fast enough that you can make the same or better income at lower rates.
Don't worry you still get to code, agents work best when you show them good quality example code.
There will be lots of work for people who can fix vibecoded messes for some time
This is what I'm leaning into, while my coworkers keep creating bigger PRs winch make the codebase worse without understanding what goes it, I took the opposite direction and focus on targeted cleanups, with proper measurements of improvement.
AI makes so easy to add metrics to quickly asses if an idea makes sense to implement, that it feels like cheating, sometimes I ask to create components form scratch with the bare minimum data simply to check if it's viable to make the "real" logic faster.
Hmmm... I can't really guarantee that this is useful as advice, but I can try my best to answer the question. I'm a member of a 6-person collective (5 engineers + 1 product manager). Being a collective is both good and bad — bad, because we need much more work to meet everyone's needs, but good, because if we lose a client, we can almost always shuffle things around so that everyone has at least some work. Also means we can share much of the admin work necessary for running a business, which is nice.
We've found success within our niches (which are mostly collaborative rich text editing and ebook/audiobook forced alignment). We write about our challenges and successes in these fields as much as we can (we do the writing, not LLMs), and share this writing everywhere we think it might be relevant. The result is that people who find that their LLMs are struggling with these fairly challenging domains are at least somewhat likely to have seen some of our writing, and either reach out to us or are receptive when we reach out to them.
I will say that we've been trying to get another client for the past couple months and it's definitely been much more challenging than it was, say, two years ago, but we have a few promising leads, still. I think that the pressure will ease up a bit as companies start to see the value that they lost in assuming they could replace domain experts with LLMs (this is happening a bit already).
But ultimately the way that we're keeping ourselves in the game, so to speak, is doubling down on our areas of expertise. We keep working on our open source projects (we have a rich text editing framework based on ProseMirror and I separately have a self-hosted book alignment platform) and trying to get them some visibility, and keep writing about the things we find interesting.
I think this is going to be the key to success for most small businesses and freelancers. Bonus points if your niche has an audience with some level of distaste for AI (the arts, as an example) - but this isn't a requirement.
You can always expand outside of your niche audience, but you can network better if you can show that you've worked with customers in a similar organization.
Agreed! The easiest sell is definitely "We've solved this problem before, here, see, look". We still do work with other clients that don't need collaborative rich text editors or aligned books (one of our longest standing partnerships is with https://movement.vote, we even have a discounted rate with them because we just like working with them), but I also personally knew the CTO and he knew I was good at my work. For connecting with companies that we don't already have a personal relationship with, being able to point to previous work in that domain is basically how we win all of our contracts.
I am part of a mini collective (it’s only 3 of us), but we don’t have enough work to spread around anymore. Most of our contracts weren’t renewed because “AI can do it cheaper”, and the ones that were being discussed didn’t sign for the same reason.
How did you assemble your collective? Was it people you already knew, or did you seek each other out?
if you overlap in skills and offering, how do you avoid competing with each other? And if you’re specialized each in different areas, how do you shift existing work around that is specialized work only one of you can do presumably?
We're probably at max capacity right now for engineering and product management, though if we can figure out how to get funding/work for them I'd really like to bring on a designer that we have all worked with.
With one exception, all of us worked together in the same group at The New York Times. Our newest member is one that only I knew, but that was a really good fit with the group. The other members chatted with him multiple times and got to know him before we invited him to join. I don't think we'll ever extend an invite to someone that at least one of us hasn't worked with extensively in the past, because our operating agreement gives everyone equal voting power and ownership of the company. So we're incentivized to both remain small and only pull in people that we trust.
Of the 5 engineers, three of us are domain experts in collaborative rich text editing, so we have the easiest time shifting work around. We try to onboard new members to our rich text editing projects when they join so that they can hop onto a project if needed, since that's where most of our contracts come from. One of our contracts is just general full stack software development though, which all of us are competent on (if only they wanted us to work with them for more hours!). If someone needs hours and we want to pull them on to an existing project in a domain they're not familiar with, we ramp up their hours on that project incrementally and have the current lead for the project spend time onboarding them and reviewing/assisting their work. One of our members joined at the beginning of this year, for example, and they only had superficial experience with rich text editing, so to start I swapped one day a week of the contract with them and spent a few weeks ramping them up. Now they're the lead on the project and I only work one day a week on it, because I have another project I'm leading.
We don't really have any issues with competing with each other, I don't think. This is also in our operating agreement: whoever most needs work/hours/pay gets first dibs on any potential new work, regardless of skillset and expertise (but none of us would put ourselves on a project that we weren't prepared to accomplish).
I love the collective idea, I wish you continued success!
Back to the OP:
I think that the pressure will ease up a bit as companies start to see the value that they lost in assuming they could replace domain experts with LLMs
Good point, at the level of the current frontier it takes some time before the agents accumulate enough technical debt to start making a mess of things in ways they can't fix. There's a threshold both in complexity and codebase size. We may not have seen the peak of the LLM disillusionment wave yet.
It's tricky though because newer models can kick the can down the road by cleaning up the mistakes of the previous generation, at least partly. And at the same time more and more developers are going to figure out how to use agents to make things that don't suck.
That won't stop people who have no idea what they're doing from continuing to churn out code no one understands though, so I think on balance the demand for experts won't go away.
Yeah I don't think that there's going to be a straight backwalk to the before times by any stretch. I do think that at least some companies that thought they could drop all of their domain experts in favor of more LLMs are and will continue to regress at least somewhat on that stance, as they start to realize how much expertise is necessary to wield these things effectively. How long that will take is obviously unclear, as is whether improvements (which, from what I've seen/experienced, have slowed considerably from, say, the first half of this year) to LLMs themselves will actually overcome the gaps that exist right now.
I've also had awkward conversations where I have to convince a prospective client that an experienced developer with AI beats them as a noob with AI. I'm exploring a pivot into a more physical/hands-on business but I'm totally happy to go back to full-time employment if need be. I'm getting plenty of invitations to apply to various companies from people in my network.
delphi | a day ago
Do the thing that AI can't do: Sit down with your clients, talk to them, figure out an exact spec, and then maybe do the visual and UX design by hand? AI's not very good at that currently, compared to a skilled professional. Customer service and design is what makes you stand out these days.
snake_case | 19 hours ago
Its exactly the same elsewhere, Ive become the goto for multiple teams because these people would rather ask me questions even when the Claude that our company pays for would answer it better most of the time.
People just dont wanna talk to bots, no matter how good the bot is, when theyre talking to a bot and not a human they feel ignored.
post_below | a day ago
There will be lots of work for people who can fix vibecoded messes for some time. You could lean into that market.
If you're a coder, learn to use AI effectively, meaning: The end result is equivalent quality to a hand written codebase. It's far slower than vibecoding, but still faster than it used to be. Maybe fast enough that you can make the same or better income at lower rates.
Don't worry you still get to code, agents work best when you show them good quality example code.
marcus-aurelius | 23 hours ago
This is what I'm leaning into, while my coworkers keep creating bigger PRs winch make the codebase worse without understanding what goes it, I took the opposite direction and focus on targeted cleanups, with proper measurements of improvement.
AI makes so easy to add metrics to quickly asses if an idea makes sense to implement, that it feels like cheating, sometimes I ask to create components form scratch with the bare minimum data simply to check if it's viable to make the "real" logic faster.
smores | 19 hours ago
Hmmm... I can't really guarantee that this is useful as advice, but I can try my best to answer the question. I'm a member of a 6-person collective (5 engineers + 1 product manager). Being a collective is both good and bad — bad, because we need much more work to meet everyone's needs, but good, because if we lose a client, we can almost always shuffle things around so that everyone has at least some work. Also means we can share much of the admin work necessary for running a business, which is nice.
We've found success within our niches (which are mostly collaborative rich text editing and ebook/audiobook forced alignment). We write about our challenges and successes in these fields as much as we can (we do the writing, not LLMs), and share this writing everywhere we think it might be relevant. The result is that people who find that their LLMs are struggling with these fairly challenging domains are at least somewhat likely to have seen some of our writing, and either reach out to us or are receptive when we reach out to them.
I will say that we've been trying to get another client for the past couple months and it's definitely been much more challenging than it was, say, two years ago, but we have a few promising leads, still. I think that the pressure will ease up a bit as companies start to see the value that they lost in assuming they could replace domain experts with LLMs (this is happening a bit already).
But ultimately the way that we're keeping ourselves in the game, so to speak, is doubling down on our areas of expertise. We keep working on our open source projects (we have a rich text editing framework based on ProseMirror and I separately have a self-hosted book alignment platform) and trying to get them some visibility, and keep writing about the things we find interesting.
ShroudedScribe | 18 hours ago
I think this is going to be the key to success for most small businesses and freelancers. Bonus points if your niche has an audience with some level of distaste for AI (the arts, as an example) - but this isn't a requirement.
You can always expand outside of your niche audience, but you can network better if you can show that you've worked with customers in a similar organization.
smores | 18 hours ago
Agreed! The easiest sell is definitely "We've solved this problem before, here, see, look". We still do work with other clients that don't need collaborative rich text editors or aligned books (one of our longest standing partnerships is with https://movement.vote, we even have a discounted rate with them because we just like working with them), but I also personally knew the CTO and he knew I was good at my work. For connecting with companies that we don't already have a personal relationship with, being able to point to previous work in that domain is basically how we win all of our contracts.
[OP] preposterous | 18 hours ago
I am part of a mini collective (it’s only 3 of us), but we don’t have enough work to spread around anymore. Most of our contracts weren’t renewed because “AI can do it cheaper”, and the ones that were being discussed didn’t sign for the same reason.
How did you assemble your collective? Was it people you already knew, or did you seek each other out?
if you overlap in skills and offering, how do you avoid competing with each other? And if you’re specialized each in different areas, how do you shift existing work around that is specialized work only one of you can do presumably?
smores | 17 hours ago
We're probably at max capacity right now for engineering and product management, though if we can figure out how to get funding/work for them I'd really like to bring on a designer that we have all worked with.
With one exception, all of us worked together in the same group at The New York Times. Our newest member is one that only I knew, but that was a really good fit with the group. The other members chatted with him multiple times and got to know him before we invited him to join. I don't think we'll ever extend an invite to someone that at least one of us hasn't worked with extensively in the past, because our operating agreement gives everyone equal voting power and ownership of the company. So we're incentivized to both remain small and only pull in people that we trust.
Of the 5 engineers, three of us are domain experts in collaborative rich text editing, so we have the easiest time shifting work around. We try to onboard new members to our rich text editing projects when they join so that they can hop onto a project if needed, since that's where most of our contracts come from. One of our contracts is just general full stack software development though, which all of us are competent on (if only they wanted us to work with them for more hours!). If someone needs hours and we want to pull them on to an existing project in a domain they're not familiar with, we ramp up their hours on that project incrementally and have the current lead for the project spend time onboarding them and reviewing/assisting their work. One of our members joined at the beginning of this year, for example, and they only had superficial experience with rich text editing, so to start I swapped one day a week of the contract with them and spent a few weeks ramping them up. Now they're the lead on the project and I only work one day a week on it, because I have another project I'm leading.
We don't really have any issues with competing with each other, I don't think. This is also in our operating agreement: whoever most needs work/hours/pay gets first dibs on any potential new work, regardless of skillset and expertise (but none of us would put ourselves on a project that we weren't prepared to accomplish).
post_below | 17 hours ago
I love the collective idea, I wish you continued success!
Back to the OP:
Good point, at the level of the current frontier it takes some time before the agents accumulate enough technical debt to start making a mess of things in ways they can't fix. There's a threshold both in complexity and codebase size. We may not have seen the peak of the LLM disillusionment wave yet.
It's tricky though because newer models can kick the can down the road by cleaning up the mistakes of the previous generation, at least partly. And at the same time more and more developers are going to figure out how to use agents to make things that don't suck.
That won't stop people who have no idea what they're doing from continuing to churn out code no one understands though, so I think on balance the demand for experts won't go away.
smores | 15 hours ago
Thanks!
Yeah I don't think that there's going to be a straight backwalk to the before times by any stretch. I do think that at least some companies that thought they could drop all of their domain experts in favor of more LLMs are and will continue to regress at least somewhat on that stance, as they start to realize how much expertise is necessary to wield these things effectively. How long that will take is obviously unclear, as is whether improvements (which, from what I've seen/experienced, have slowed considerably from, say, the first half of this year) to LLMs themselves will actually overcome the gaps that exist right now.
moonwalker | 19 hours ago
What does this mean?
[OP] preposterous | 18 hours ago
Fiver, malt, rent a coder etc where you compete for projects by offering a lower rate than the other people doing the same rate lowering.
moonwalker | 11 hours ago
Ah, ok. Thanks for clarifying
teaearlgraycold | 13 hours ago
I've also had awkward conversations where I have to convince a prospective client that an experienced developer with AI beats them as a noob with AI. I'm exploring a pivot into a more physical/hands-on business but I'm totally happy to go back to full-time employment if need be. I'm getting plenty of invitations to apply to various companies from people in my network.
loremipsum | 5 hours ago
Well, if any of your friends is looking for a senior front end engineer, 👋🙂
teaearlgraycold | 4 hours ago
I think everyone is currently hiring for "person that will work in a wide-variety of software domains as a Claude-assisted idiot".