Is AI going to replace software developers?
The honest answer from someone who runs a 39-project portfolio with these tools in daily use: the job is being restructured rather than removed, some rungs of the ladder are genuinely disappearing, and the scarce skill has shifted from producing code to judging it.
Asghar Ali · Chief Technology Officer, Kakushin.io LTD
I get asked this by engineers on my team, by clients deciding whether to hire, and by people deciding whether to study computer science at all. I run technical direction across thirty-nine projects with these tools in daily use, so I have some evidence rather than just an opinion.
The short version: no, and also the job is not going to look like it does now. Both halves of that matter, and most takes only carry one.
What the tools genuinely do
I want to be precise here, because both the hype and the backlash are imprecise.
Model-assisted development has changed my throughput materially. Boilerplate is gone. Unfamiliar APIs take minutes instead of an afternoon. A first draft of almost anything appears instantly, and a bad first draft you can edit beats a blank file. On our estate there is an automated pipeline that sweeps pull requests, reviews, fixes, merges, deploys and announces — built on an agent SDK, and it does real work every day.
Anyone telling you this is a toy has not used it seriously.
What has not moved
The bottleneck was never typing.
The expensive parts of this job are deciding what to build, knowing whether it actually works, and living with it for three years. A model collapses the cost of producing code and leaves those three exactly where they were.
Watch where the queue forms. On one of our products the client's founder — not a professional engineer — ships her own pull requests, and the delivery constraint on that project is my review capacity, not her output. The work did not vanish. It moved to the person who has to say "yes, this is correct."
The part people avoid saying
There is a real and uncomfortable consequence: the bottom rung is being sawn off.
A meaningful share of what junior engineers were historically paid to do — small well-specified tickets, boilerplate, glue, test scaffolding — is exactly what generation is best at. The tasks that used to build judgment through repetition are the tasks now being automated.
That is not "AI replaces developers." It is worse in a specific way: it damages the pipeline that produces the senior engineers everyone will still need. If you only hire people who can already review generated code competently, you are consuming a supply you have stopped replenishing.
I do not have a clean answer. What I do is deliberately keep some work hand-built by people who need to grow into it, and treat that as a training cost rather than an inefficiency.
Where the value actually moved
If you are deciding what to be good at, aim here:
Judgment about correctness. Not "does it run" but "is it right, under load, with real data, for the case nobody described." This is the scarce skill now and it is getting scarcer.
Systems thinking. Models produce locally plausible code. They do not hold your architecture, your failure modes or your migration history. Someone has to, and that someone is increasingly valuable.
The boundaries. Authorisation, tenancy, spend ceilings, what may leave the network. These look like code but they are policy, and policy is not inferable from the surrounding source.
Debugging what you did not write. Increasingly the code in front of you was generated. Reading unfamiliar code under time pressure, forming a hypothesis and proving it — that skill just became central.
Talking to people. The requirement lives in a person's head and is usually half-articulated. Extracting it remains entirely human work.
What I would tell someone starting out
Learn the fundamentals properly — networks, data structures, how a database actually behaves — because they are what let you tell good generated code from plausible generated code. That discrimination is the whole job now.
Use the tools from day one. Refusing them is not integrity, it is just slower. But make yourself understand every line you ship, and when you cannot explain why something works, stop and find out. The gap between "it works" and "I know why it works" is precisely where your career is.
The honest summary
The people at risk are not developers. They are developers whose entire value was producing code to a specification somebody else wrote — and that was already a precarious position before any of this.
The job is turning into something closer to editing than authoring: less production, more judgment, more responsibility per line shipped. That is a harder job, not an easier one, and it is going to need more experience rather than less.
Which is the part that worries me. Not that AI replaces developers — that we stop training the ones who will have to supervise it.
Related: why human validation is still the constraint and the security risks nobody checks in vibe-coded apps.