AI debate points to 30% discounts and fewer programmers

AI reaches code: 30% discounts and fewer junior staff

The first symptom is not a layoff: it is a discount. Companies have reportedly started demanding an upfront 30% reduction from software consultancies, even before knowing how much of that work artificial intelligence will absorb. Less margin for payrolls, more processes brought in-house. The adjustment is called productivity; the person signing it usually reads it as a pay cut. And the profession that spent years complaining that technical quality was not being paid suddenly finds itself debating whether it needs to exist at all.

The cut already being demanded in software contracts

The sequence that repeats most often is always the same: the company asks for the discount before knowing what it will automate, the consultancy squeezes its subcontracted staff, and whoever is at the bottom of the table is left out. There is a name for this: the Uber moment. A technology that lands in a sector intensive in time and turns the business upside down.

There are detailed cases: in one large department, external professionals with a minimum level of skill were reportedly brought in-house; for the rest, their consultancy was simply told their services were no longer needed. The return to in-housing is most visible in banking, where in-house staff regain weight while external contracts lose volume.

The trigger is not just efficiency. According to this reading, companies have realised that their technology department is about to fall into the hands of the big cloud providers and automatic code generation tools, and they prefer to keep in-house whoever supervises that process.

Can artificial intelligence program on its own without human supervision?

It cannot carry a complex development through, but it does sweep away the part that consumed the most hours. Transcribing hundreds of programs, converting variables, inserting statements, documenting legacy code or regression testing when deploying new versions are mechanical tasks where the machine makes fewer mistakes than the tired human. The thread suggests that a platform migration —a huge market— could be automated up to 95%.

There lies the goldmine and there lies the hole. A client who does not know what they want, a request nobody can validate, an impossible requirement that has to be dismantled with arguments: none of that is solved by a text generator. The most repeated example is the absurd, immediate commission: a big-budget video game for tomorrow. The machine, at best, returns a tic-tac-toe copied from a tutorial.

A conclusion repeated throughout the thread is uncomfortable for the profession: expert knowledge is still necessary, but for a fraction of the sector. The code of veterans is documented so that dependence on those people disappears, and the tool's output is reviewed. Reviewing is faster than writing, and that changes every calculation.

From 14 interns to 3: the examples cited outside the sector

In sectors where automation arrived earlier, the debate already handles concrete examples. A large law firm reportedly went from two senior lawyers with 14 interns to three seniors with three interns, with the same output —or better— thanks to the tools. In the courts, the example being discussed is five judges resolving 3,000 rulings a year versus two judges handling 10,000.

It should be taken for what it is: a scenario defended in the analysis, not an official statistic. But it points to a concrete pattern. What falls is not the whole job, it is the part of the pyramid devoted to repetitive, low-judgement tasks. And that is precisely where entry-level contracts were concentrated.

The two competing narratives: end of the profession or demand lever

One current holds that history repeats itself and pays off. Nineteenth-century textiles clothed people who previously wore little more than a blanket; calculators and spreadsheets did not end accountants, they demanded more of them and gave them more production capacity. Translated to software: if ten things were done before, a hundred will be done, and what could not be afforded before now can be.

The opposite version is harsher. The sector would have spent two decades simplifying its work to sell it cheap, and that simplification is exactly what is now being automated. The thread points to 30% of the sector lost within five years; another, more drastic message speaks of 90% fewer professionals. Whoever has experience and judgement holds on; whoever was starting out is left without an entry door, which is the problem nobody knows how to fix.

What if artificial general intelligence arrives?

When the debate stretches forward, the scenario changes in nature. A general AI capable of doing 99% of tasks better than a human, including creative ones, would stop being a tool of the trade and become its replacement. That horizon drags in discussions about universal basic income, cheap leisure and even a change in the very concept of money.

Meanwhile, the fight is more down to earth: who pays for training, who keeps the margin and what happens to those starting out. The comfortable timeline that was given —six to eight good years to train, push hard and save— may be exactly the time it takes for the door to close. Does the margin the machine frees up end up in the payroll of whoever operates it or in the pocket of whoever buys it?