Every engineer I know is trying to work out where they fit now. One camp says engineering is dead. The other says fundamentals matter more than ever. I don't think either camp has it quite right, and I want to put down a prediction, so I can reflect on it later.
The bar has moved
Things that used to need an engineer don't anymore. A personal website, a small web app, an internal tool: someone with an idea and an AI model can build it themselves, and it will comfortably handle the load that a personal site or small business actually sees. Tell the model "my site is slow" and it will go and find out why.
That's not a toy. The models lay down a genuinely decent foundation. How far a non-engineer can take an idea, without real engineering practices and without an engineer looking at it, has gone up a lot, and it keeps going up.
And it isn't only the models. People are now sharing basic agentic harnesses: starter kits with the conventions, checks and guardrails an engineer would have put in place. A non-engineer can pick one up and get a sensible setup on day one. Depending on what they're building, they may never need an engineer at all. The harness raises the floor.
The "we'll just do more" argument
The optimistic counterargument is that engineers won't shrink, we'll just do ten or a hundred times more.
I don't buy it, and the reason isn't technical. Maybe you can now engineer a hundred ideas. But someone still has to hold each idea in their head, see it through, put it in front of the market, watch what happens, and iterate. That's product work, and it runs on human attention.
My guess, and it is a guess, is that a person can take on maybe five times more than before. Not a hundred. The build capacity has exploded, but the thinking capacity hasn't. So demand for engineering doesn't grow to match supply. It shrinks.
Engineers as a luxury item
Here's where I think it goes. Having full-time engineers becomes a luxury. Only the biggest companies, the ones that can afford to overspend or that run at genuinely critical load, will keep them in-house. And even there, the numbers will come down.
Think about cars. Almost everyone drives, but almost nobody employs a mechanic. Only a company with a big fleet keeps mechanics on staff. Everyone else takes their car to a workshop.
And plenty of people will happily drive a car with a rattle or a warning light on, as long as it gets them where they want to go. Software will be the same. A lot of products will run with faults nobody ever fixes, and that will be fine for their owners.
The workshop model
Agencies aren't new. Workshops have always existed, and plenty of engineers already work in them. What I think changes is where most engineers sit. Today the majority work in-house on a single product, and agencies are the smaller side of the industry. I think that flips.
We're going to see a lot more agencies, and engineers will centralise into them. Most will stop working on one product and start working like mechanics.
Someone brings their product into the workshop. The engineers look at where the AI keeps getting stuck, tune the agentic harness, fix the problems, put the current best practices in place, and send it back out. Then they might not see that product again for a year.
And then, like a car, it comes back for its annual service. The engineers check it's still following best practice, talk through the problems the owner has hit, and bring it up to date with whatever the industry has learned since.
This is where the shared harness comes back in. The workshop is for when the off-the-shelf harness stops being enough, and the annual service is where you swap in the latest harness.
And it works the other way too. Every repair teaches the workshop something. When the same fault keeps coming through the door, the engineers fix it in the harness itself, the way car makers learn from what garages keep seeing and fix it in the next model. A workshop that sees hundreds of products learns faster than any single team or any static starter kit, so its harness stays ahead.
The analogy isn't perfect. Software changes faster than cars do, so I'd expect plenty of these relationships to be retainers rather than one-off visits: a mechanic you can call, not just one you book. But the shape is the same. The engineer is outside the business, serving many clients, and the business model is the agency model: your income depends on how many cars come through the door.
Yes, some people will never bring their car in. But that lower starting point also means far more drivers than before. More people building products means more products that eventually hit something the harness can't handle. So in a sense it balances out: fewer in-house engineers, but a lot more cars on the road needing the occasional mechanic.
What this means for engineers
If I'm right, the question for an engineer isn't "will AI replace me". It's "do I want to work in a fleet garage or a high street workshop". The skills that matter shift towards diagnosis, towards knowing current best practice across many codebases, and towards being good at getting a stranger's product healthy quickly.
That's my prediction. I could be wrong, and I'd like to hear why. But I'm looking at what people around me can now build without me, and this is the picture I keep coming back to.
Top comments (0)