The teams are already getting smaller
If you’ve been anywhere near the job market over the last year or two, you already know the teams are getting smaller. I’ve watched it happen at the big-company level, and I’ve watched it up close with friends who own businesses and employ small teams, some of them quietly trimming down and a few of them replacing whole functions altogether.
What gets me is how normal it used to be to see a 20-to-30-person team behind an app that was genuinely successful, pulling in millions on the store. I saw teams that size all the time, and now I see them less and less. What I see instead are much smaller teams, or in a lot of cases founder-led ones where the founder just takes on whole functions themselves, or these very lean setups where each function is basically a single specialist and they all collaborate together. And it’s happening because AI is taking on more and more of the work those bigger teams used to do.
At the same time I’m seeing the other side of it: individuals building their own little apps and internal tools for problems they’d previously have paid a monthly SaaS subscription to solve, and companies of every size starting to pull away from third-party tools and just build the thing themselves. Even the people who aren’t technical enough to build for themselves have never had more ready-made options to pick from. It all points in the same direction.
Look how far AI has come in the last three years
I don’t think we’ve really stopped to take in how fast this has moved. Cast your mind back to when ChatGPT first went mainstream, which was barely three years ago. It could hold a conversation for a little while and then lose the thread completely, start hallucinating, quietly drop in quality until you gave up and started over. That was the state of the art at the time, and it was genuinely impressive. Now the same line of tools is building SaaS products, shipping mobile apps, and in some cases running the guts of businesses turning over millions. I use this stuff every single day to build things, and I still find that leap hard to hold in my head.
Now, I’m not naive about where it actually is today. It still does daft things when you don’t guide it properly. My wife runs marketing for StillMind, and one of the things she’s building is an AI setup that generates meditation videos for YouTube and gets them localised into different languages. Somewhere in that build the AI decided the right place to store the project’s environment variables was in her .zshrc file instead of a .env file, which is the kind of thing no developer would ever do, and because she’s a non-technical co-founder, not an engineer, there was no way she’d have caught it. I had to step in as the technical one and untangle it.
That’s the current reality of working with these tools: they’ll happily make a silly call and carry on with total confidence, and it takes someone who knows better to notice and put it right.
But here’s the thing. The distance between where it is now and it not doing that stuff anymore feels tiny to me, next to the distance it’s already covered in three years. So when people point at the silly mistakes as proof our jobs are safe, I just don’t buy it. It honestly doesn’t seem far-fetched that within a couple of years even non-technical people will be building genuinely excellent products with these tools. And if that turns out to be true, the real question is how the software industry as we know it survives it.
The jobs that are left, and why
My honest guess is that it doesn’t survive in the shape it’s in now. It splits into two.
Most of the delivery work, the actual building of the average product, gets less and less valuable over time, because the AI does more and more of it to a good enough standard that you can’t really justify a team of experienced people for it. And if I’m being honest, the vast majority of what software actually is sits right in that bucket: the productivity apps, the self-development tools, the endless SaaS, all the everyday stuff. For most of it you either won’t need an experienced developer at all, or you’ll need a single one where you used to need a whole team, and I think we have to be straight with ourselves about that rather than pretend otherwise.
It’s a big part of why the market feels the way it does right now, because there are so many developers out there that the roles still going are getting brutally competitive, simply because there are far more developers than there are jobs left for them.
What survives, and actually gets more valuable, is the other kind of work: the small number of things that are so critical, so complex or so high-stakes that we simply can’t hand them over. And I want to be careful here, because it isn’t really about competence anymore. It’s about trust. Think about air traffic control, or medical systems, or the people building the AI itself. Even if the AI got brilliant at those, good enough to run them on its own, I don’t think we ever get to a world where we just let it, with no human anywhere in the loop, on the things where one mistake is catastrophic.
Those roles stop being about someone who can do the work and become about someone capable enough to verify it, to automate around it, and to be the accountable human standing behind it. There’ll be far fewer of those jobs, but each one matters enormously.
It’s a bit like the 80/20 rule tipping over into diminishing returns across a whole industry. It won’t all be strictly automated, either. It’s more that the routine 80% either gets automated or gets built away by a new kind of builder, and what’s left is the difficult, high-skill, high-trust 20%. Very valuable, and very hard to get into.
The rise of AI builders
There’s a second thing that survives, though, and it’s a more hopeful one: a whole new kind of builder.
I think we’re heading towards a world of AI builders, people who build things with AI on behalf of everyone who doesn’t want to, or doesn’t have the time to, do it themselves. And here’s the part I find genuinely interesting. I don’t think they’ll need the deep, ground-up knowledge of how the software actually works to get a great result out of the AI. It’s a little like the WordPress era, when the CMS made websites less technical and suddenly there were “WordPress developers” who were really theme-and-plugin people working mostly through a UI. I lived through the middle of that one, my first real product success was a payment plugin I built for exactly that world, so I’ve seen how a tool can quietly redraw the lines of who counts as a builder.
But I think this time the comparison undersells it, because the outcomes people will get from directing AI well are going to be miles beyond anything the old WordPress-and-plugins setup could ever produce.
Looking unique is going to be the hard part
If everyone can build, then building stops being the thing that sets you apart. And I think that’s already happening.
AI can absolutely make something that looks good, all day long, and that’s exactly the problem. It’s just not something many people realise, or spend the time learning how to prevent. If you just let it do a good enough job and ship whatever comes out, you get something that looks credible on the surface but looks like every other AI product out there, or feels ever so slightly off, or has something not quite right about the quality when you sit with it. You can see it everywhere already: a wash of sameness, the same phrasing, the same layouts, the same content, all blending into one.
So the hard part isn’t taste, because the AI can do tasteful. The hard part is making something that genuinely stands out and feels like yours when everyone around you has the exact same tools. That takes real time, real effort and real skill, learning not just how to get AI to deliver something, but how to get it to deliver something unique. The people who put that work in, and keep putting it in, are the ones who’ll get noticed in all the noise. The people who don’t will quietly disappear into the wash with everyone else.
It won’t stop at software
I don’t think any of this stays inside software, either, and this next bit is pure speculation on my part, so take it as that.
Robotics has come a long way, and my read is that the biggest thing holding it back has always been the software, the intelligence to actually cope with a messy, unpredictable physical world. That happens to be exactly the thing AI is about to be revolutionary at. So I don’t think it’s unrealistic that fairly soon we see robots taking on real physical labour: the cleaning, the basic construction, the repetitive manual jobs. Which means this isn’t a story about the tech industry getting disrupted while everyone else watches from a safe distance. It reaches a lot further than that, into plenty of livelihoods that have nothing to do with code.
We get to be human again
So I’m holding two things at once, and I’ve made my peace with not resolving them. The next few years are going to be genuinely hard. The job market is already brutal, with far too many good people fighting over each role, and I think it gets worse before it gets better. There’s real money to be made right now by the people building solutions, at least in the short term, but the hard part there was always getting discovered and actually earning from it, and that doesn’t get any easier when everyone can build. The shape I’d bet on for the next few years isn’t companies fading away. Plenty of companies are going to thrive. What I think we’ll actually see is companies getting leaner, everything from small teams to massive corporations doing more and more with fewer and fewer people, and alongside that, more and more individuals going it alone. A minority will do very well out of it: the ones who are entrepreneurial and build early, and the ones who join the exciting AI-first teams. But a lot of people are going to struggle, and genuinely suffer, because all that downsizing across the board leaves far fewer jobs to go round. I don’t want to wave that away.
But when I let myself look further out, I get genuinely excited, and it comes down to my kids. For the last fifty years or so we’ve slowly been asking people to behave more and more like computers: sit at the machine, take the inputs, follow the process, produce the outputs. The work got less human, not more. What I think AI actually does, if we can get through the painful part of it, is hand the computer work back to the computers and give the human work back to us.
I feel it most in my own field. So many product teams skip the genuinely human part of the job, the research and the validation, actually sitting with users and confirming that what you’re building matters to them, and jump straight to a founder’s gut feeling instead. That part was always the most important, and it was often the first thing cut when time got tight. In this new world I think it finally gets the time it deserves, because it’s the bit the AI can’t do for you, and it becomes the difference between the products that land and the ones that don’t. People get to be human because they get to go and talk to people again.
And I think it shows up well beyond work. A better balance, maybe shorter days, maybe fewer of them in the week. That might be optimistic, and maybe something more structural like universal basic income ends up being part of the answer, I genuinely don’t know. But I don’t think it means the end of a working economy so much as a return to an older version of one.
Go back far enough and every town had its blacksmith, someone who’d gone deep on a real skill and made their living from it. Give a lot of people a lot of their time back and I think you get a modern take on exactly that: the local running enthusiast becomes the town’s running coach, the bushcraft obsessive gathers a little community around them. And it won’t all be physical, either. Just as much of it will play out online, in the gaming communities and the countless niche corners of the internet where people already gather around a shared obsession. Either way, people earn through communities, hobbies and hard-won skills rather than through a job that treated them like a slightly unreliable machine. More social, more human, more real.
I know that’s a utopian way to see it, and I know I’m probably being too optimistic. But I can’t really help it, that’s just who I am, and it’s the same instinct that keeps me wanting to build things in the first place. The transition is going to be hard, and a lot of people will get caught in it, and none of the hopeful stuff is me trying to wave that away. I just genuinely believe that if we can get through the difficult decade in front of us, the computers finally doing the computer work means the rest of us get to go back to being human. And honestly, that feels like a better world to hand my kids than the one we’ve got.