The tool amplifies whoever holds it

This is not a technical post, rather a snapshot of the trade as I watch it go by, year after year, in the job ads and in the discussion threads. I am going to talk about two things that look unrelated at first glance: what recruiters ask for and what AI does to junior profiles. By the end, it is the same subject.

1. The five-legged sheep, at the price of an ordinary one

Some people genuinely manage to wear several hats at once, with weaknesses of course, one leaning more toward infra, another more toward dev. I, for one, can read, understand, adapt or fix code and an architecture across many idioms, but exploiting a framework like Laravel, Symfony, Django or Flask, churning out business features and validating PRs all day long, that is not for me. It is not my ground and I would rather say so than fake it.

These swiss-army profiles often have an enormous capacity to ramp up and they match the ads without straining too much. Except that recruiters want them to tick every box at once: advanced Kubernetes expertise, plus dev in this or that language or framework, plus the rest, otherwise it is dead before the interview.

And I am not even talking about the cut-rate salary yet. Go on, candidate, build us an infra just so, take care of industrialization, melt our bills, replace two or three roles on your own and while you are at it bring the coffee, but for 65 or 70k, or a 500 euro day rate. They are hunting a five-legged sheep and paying it at the flock’s price.

That such profiles exist, nobody doubts. That they will accept, over the long run, being paid at that price while the roles pile up on them, is another story and that is where the market lies to itself.

2. What we are manufacturing inside the trade

I will be told that AI is going to democratize access to knowledge and close the gap. On that point, I do not share the ambient optimism. I already saw a fracture and it is widening instead of closing.

The worst is no longer even those who never touch code. It is what we are manufacturing inside the trade itself: a whole generation over-specialized, trained on an assembly line, glued to its idioms and frameworks, shipping code without ever having looked at what lies underneath. They do not know what a syscall does, how a TCP/IP packet flows, what happens below the abstraction they handle all day and above all it does not interest them for a second.

This is not an insult, it is an observation and the product of a training system that optimized immediate employability at the expense of fundamentals. We taught people to plug bricks together without ever showing them the inside of the bricks.

3. The tool amplifies, it does not correct

Put an LLM in those hands and it does not lift them. It lets them slide faster down the same rut.

That is the mechanic I take away from these last two years: the tool amplifies whoever holds it. A handful of profiles will use it to dig, to verify and to genuinely learn what lies underneath. Many will use it to think in their place and ship slop by the yard, masking the hole a little faster, a little more cleanly, until the day the hole reminds them of itself in production.

A curious junior, who asks the model why rather than what, who recompiles the explanation in their head and tests it against reality, comes out of it grown. A rushed junior, who copy-pastes the plausible answer and moves to the next ticket, comes out at exactly the same level, just with more lines to their name and a slightly better-hidden debt. The same tool, two opposite trajectories and the difference does not come from the tool.

4. The bottleneck is discernment

So what I would put on the table is not one more skill to tick in an ad, it is this question: who uses the tool to know and who gets served by it.

The bottleneck was never the machine. It is in discernment and in the fundamentals we no longer want to learn because they pay nothing in the quarter you learn them. An LLM speeds up code production, it does not speed up the forming of judgment and conflating the two is the casting error I keep seeing repeated.

That is also why I stay cautious about how we recruit and evaluate in the agentic era. I detailed elsewhere why a degree or a certification does not measure judgment and how I prefer to have someone code and then recount a lived incident rather than tick boxes; that is the companion post to this one, Agentic LLMs in production . Here I will not replay that argument, I only point to the common root: we try to buy stacked skills when the real asset, rare and slow to build, is the capacity to understand what one is handling.

A tool never saved anyone

• The market wants the five-legged sheep (Kubernetes expert, multi-framework dev, industrialization, cost reduction) and pays it at the price of an ordinary sheep, 65/70k or a 500 euro day rate. It does not hold over time.
• The real fracture is not between those who code and those who do not, it is inside the trade: a generation trained on an assembly line that no longer knows what happens below the abstraction and does not care.
• AI does not lift these profiles, it amplifies them. The curious one digs faster, the rushed one hides the hole faster. The tool does not choose the trajectory.
• The bottleneck was never the machine, it is discernment and the fundamentals we no longer want to learn.
• Invest in fundamentals before investing in tokens: syscalls, TCP/IP, what lies under the brick. The rest can be bought, that cannot.