A 12-month window that has expired
A year ago, model vendors were announcing in the press that their own tools would do most, if not all, of a software engineer's work within 6 to 12 months, and that half of junior white-collar roles would be wiped off the map in 1 to 5 years.
It is now August 2026. The window has expired. Where do we stand?
Software engineers are still here. They use AI to produce faster, and end up spending more time than before rereading what comes out and making it their own. Demonstrations of constellations of hundreds of agents keep circulating, and I still have not understood what they are for in production — if anyone has the answer, I am listening.
What I see after 2 years of field work is more mundane and more interesting. AI does not remove the job. It moves value towards what cannot be generated: arbitrating, owning a decision, saying no to a sponsor, holding a disagreement in a room where 8 people hold 8 truths.
I thought my own job was about to disappear
I am not speaking from a distance. 2 years ago, I understood that generative AI was going to end the job I was doing. In 2 minutes, it delivered what had taken me 10 years to build. The demonstration was humiliating, and it was accurate.
So instead of watching it work, I put it to work.
An agile coach covers 1 LPM or 1 ART. That is the standard, nobody disputes it. Over the past 2 years, I have held 1 LPM and 3 ARTs. 4 times the scope, with a level of impact higher than what I used to produce on a single one.
I am not going to play the reassuring card for all that, because the usual conclusion drawn from this is wrong. It is not the machine that held 4 scopes, it is me. It stopped making me lose time on the mechanical part, and that is all it did.
What AI genuinely disrupts is the cost of an idea. What used to take me 3 days of workshop now fits into 40 minutes. Trivial tasks and part of the support functions will migrate to algorithms, and good riddance. The time recovered goes where the expensive decisions are made: relationships and strategy.
The dividing line is not the one being announced
Developers are not going to be replaced by AI. They are going to be replaced by the developers who use it.
This is not a threat, it is a dividing line being drawn right now between those who pivot onto the new skills and those who wait and see. Speed and quality are no longer negotiable: they are chosen.
The same logic applies one level up, and it is harsher than it sounds. An organization that does not take the generative AI turn does not stagnate. It runs the risk of dying. That is not a metaphor: the world is accelerating and change asks no one's permission, it merely waits to see who follows. The question is therefore not whether your teams are ready, but whether they have decided to be.
Why the manager holds, and on what condition
One case resists every replacement prediction, and not for the reason people assume. AI will not replace managers — not because their status protects them, but because below them, there are humans.
A dashboard can assign a station on a production line. It does not detect the weak signal of a burnout starting, the family situation weighing on the last three weeks, the moment when you simply have to listen without solving anything.
What a manager really does is not managing the how. It is holding the collective together when the pieces of the puzzle want to scatter. That organic part does not delegate.
The nuance that matters: this reasoning protects the function, not the person currently holding it. A manager whose day is made of synthesis, control and validation will see their ground erode fast — I have written elsewhere about why AI transformations fail because of managers. The one who holds is the one who has already moved their value towards what remains: judgement and connection.
What I suggest doing with this period
There is an asymmetry worth naming. The replacement narrative has considerable budgets behind it and a direct commercial interest in sustaining it: fear sells licences just as well as enthusiasm does. Facing it, the augmentation narrative has only one resource, but it is a better one — evidence.
So pin this observation and bring it back in January 2027, when the next 12-month window has expired in its turn. Until then, the most useful contribution is not one more op-ed. It is to name, inside your own team, the person who has progressed the most thanks to AI this year, and to tell their case in one sentence. One concrete case beats ten analyses.
The future belongs neither to the human alone nor to the machine alone. It belongs to the augmented human: the one who delegates the mechanical and spends their energy on judgement, empathy and meaning. That is what I call Augmented Agility, and it is the only side I am defending here.
The Great Replacement will not happen. The Great Reinforcement will — provided we stop waiting to see.