Cheap Code, Expensive Judgment: What AI Does to Developer Jobs

The cost of writing code is falling toward zero. The cost of knowing which code is worth writing is not.

That gap is the whole story. Value does not disappear when execution gets cheap, it moves. It moves to the judgment that decides what to build, what to leave unbuilt, and whether the thing the machine just produced is actually correct. The PC did this. The internet did it. The cloud did it.

AI is next in the line.

You probably read a louder version of this on your way to work. “AI Will Replace Software Engineers.” “Why Your Next Hire Might Be a Prompt.” If one of those put a knot in your stomach, fair enough, something real is moving. But the headline asks the wrong question, and a wrong question buys you either false comfort or paralysis.

Neither one helps you.

The question itself is the trap

“Will AI replace me?” is binary. Ask a binary question about a gradual transition and you get a useless answer. Decide you are safe, and you stop adapting. Decide you are doomed, and you freeze. Either way you stand still while the ground moves under you.

Two exits tempt people here, and both feel like shelter.

The first is the Dismisser. I hear this one at every kitchen table and every architecture review:

“Come on. Every couple of years something is going to replace us and it never does. Right?”

Half right. You have heard it before, that part is true. That it always blows over is not. Ask anyone who racked servers for a living what happened to the part of the job they were best at.

The second exit is the Paralysed. Everything is changing too fast, no idea where to start. Fast, yes. Powerless, no.

So, honestly: which one of those is you? Most people I talk to are one or the other and will not say so out loud.

There is a better question, and it points forward. What does my job look like in three years, and am I moving toward that version or away from it? It grants that something real is changing, so it is not denial. And it puts you back behind the wheel.

AI replacing your job is a headline. AI changing your job is a fact. Only one of them asks anything of you.

What history actually does with a threatened profession

It pushes the profession upward and kills a task inside it. That is the whole pattern, and it has held through three transitions inside living memory.

Mainframes to PCs, the 1980s. IT departments braced for irrelevance as computing walked out of the glass house. Instead, IT ballooned. The work went from babysitting one mainframe to wiring up a thousand desktops. The task died. The profession grew.

The internet, late 1990s. Every knowledge worker felt the draft: travel agents, journalists, stockbrokers. Some of those categories did shrink, genuinely. Software development grew, because somebody had to build the thing doing the disrupting.

Then the cloud. Sharpest example of the three, and the one I sat inside. Azure and the rest were going to end the system administrator. They ended exactly one thing: racking servers in a basement. The people who moved became cloud architects, platform engineers, SREs. The people who kept arguing about whether the cloud was even real became the cautionary tale you tell junior engineers.

Every wave kills a task, not a profession. The floor rises. You rise with it or you find yourself standing on air.

History does not repeat. Right now it is rhyming, and it is loud.

Get this distinction right and most “AI replaces engineers” arguments come apart in your hands.

The machine writes code that looks right, fast, faster than any human. Prototypes, throwaway scripts, boilerplate, test scaffolding, a first pass at review. That is a genuine productivity gain and you should be taking it. But “looks right” is not “is right, at scale, in production, against the architecture and the constraints you already have.” Somebody has to read what the model produced, catch the thing it got subtly wrong, and own it.

That somebody is an engineer.

Judgment does not transfer. The model does not know your business constraints. It does not know what not to build. It will not warn you that the feature your stakeholder is thrilled about buckles under load in six months, or that it quietly breaks a compliance requirement nobody bothered to put in the ticket. The job was never the typing. It was understanding a problem well enough to model it in code, and that gets more valuable as the typing gets cheap.

Yes, I am an architect telling you that architecture-shaped judgment is the valuable bit. Note the incentive and discount me accordingly.

Most of the anxious headlines collapse the same distinction. Pulled apart:

The claim	The kernel of truth	What is actually happening
“DevOps is dead”	The siloed translation role is disappearing	The discipline is more embedded than ever. The people doing it become platform engineers and SREs
“Vibe coding replaces engineers”	AI handles prototypes and MVPs fast	Someone still has to direct, review and own production code. That is the engineer
“Junior devs are obsolete”	Companies are hiring fewer juniors, AI does the scaffolding	The on-ramp is changing, not ending. Building the new on-ramp is an open problem
“AI will write all code in two years”	AI will write more code, faster, per human	AI executes. It has no judgment about which code is the right code

Same confusion in every row. Execution mistaken for judgment.

So what is genuinely shifting

The price of execution, and with it the skills the market pays for. Be honest about this part, because pretending the job stands still is its own flavour of denial.

The floor is rising. What one developer is expected to produce, and how fast, sits permanently higher now. A developer who refuses AI tooling today looks like a developer in 2010 who refused Stack Overflow on principle. You can take that stand. It will cost you.

Teams will run leaner. Not because developers are obsolete, but because a developer with good tooling ships more and companies will do that arithmetic whether you like it or not. The skill mix moves with it. Pure coding, as a commodity, is worth less. System thinking, domain knowledge, the knack for turning a messy business problem into a coherent technical model, worth more. The stakeholder dynamic shifts too, because non-technical people can now assemble things that look like software, which makes the engineer who can judge whether that thing actually holds up more valuable, not less.

None of that ends the job. All of it changes the job. Exactly what the pattern predicted.

Adapt in three layers

Use the tools on real work. Double down on the judgment the machine cannot fake. Widen your view into the domains next door. One concrete action per layer, each one startable this week.

Layer 1: stop avoiding, start using. Real work, not toy projects. Hand the machine the repetitive execution: boilerplate, test scaffolding, documentation, a first-pass review. Take that time back for judgment. You cannot hold an informed opinion about AI in your own field if you have never seriously used it. This week: pick one AI tool and put it on real work for 30 days.

Layer 2: double down on what it cannot do. Find the parts of your work that need real judgment, context, domain knowledge. Invest there on purpose. The most defensible thing you own is the translation between business reality and technical possibility. This week: on your next technical decision, write one paragraph on the business reasoning behind it. Not for a stakeholder. For you. Build the muscle.

Layer 3: widen the aperture. The people who came through the cloud transition well understood more of the system: the business, the cost model, the operational reality. T-shaped is table stakes now. Pi-shaped is better. This week: pick one adjacent domain (security, FinOps, system design, product thinking) and give it 20 minutes a week for three months.

Frequently asked questions

Will AI replace software engineers? No. AI is automating the execution layer (boilerplate, scaffolding, first drafts), not the judgment layer. The profession is transitioning the way it did through the PC, internet and cloud waves: specific tasks disappear while the role climbs toward problem modelling, review and ownership.

Are junior developer roles disappearing? The traditional on-ramp is genuinely changing. Companies are hiring fewer juniors because AI handles much of the scaffolding that used to be the entry-level path. The learning itself cannot be handed off: debugging a strange failure at 2am, building the intuition for why systems break. How the industry manufactures those reps in an AI-assisted world is an open, unsolved problem, and it is worth naming honestly.

What skills should developers invest in now? Judgment. System thinking, domain knowledge, turning ambiguous business problems into coherent technical models. Those are the skills AI lacks, and they grow more valuable as execution gets cheap. Pair them with real fluency in the tools. You cannot direct what you have never used.

Is “vibe coding” a real threat to the profession? Vibe coding is real, and useful, for prototypes and scripts and MVPs. It is not a threat, because it produces code that looks right and someone still has to confirm it is right under production constraints and carry the responsibility. The job moves from writing every line to directing, reviewing and owning what gets written. Different work. Not less of it.

The short version

“Will AI replace me?” is a trap, because it is binary. The question that pays off: what does my job look like in three years, and am I moving toward that version or away from it?
Every technology wave kills a task, not a profession. Mainframes, the internet and the cloud each killed specific work while the profession around it grew.
AI is excellent at execution and has no judgment. It does not know your business constraints, what not to build, or the compliance requirement nobody wrote down.
The floor is rising, and it is staying up. What one developer can produce, and how fast, is permanently higher.
Adaptation has three layers: use the tools on real work, invest in the judgment AI cannot do, widen into adjacent domains.

The cloud did not kill system administrators. It killed the ones still arguing about whether the cloud was real.

“Will AI change my job?” was never the question, that was always a fact. “What am I doing about it?” is the question, and that one has an answer. Execution toward zero, judgment toward the top. Move to the version of your job where you own the judgment, and the headline stops being about you.