The Perfect Employee SAP Describes Isn't Born. It's Made.

I've been reading SAP's AI best practices for BTP, the agent section. It's a well-made document, clear and fairly honest.

The whole explanation rests on a metaphor that sounds harmless:

An AI agent is like a good employee. It observes its environment, understands the situation and acts to achieve a goal.

The difference from traditional automation, the manual says, is the same one between a factory machine and a knowledge worker (someone like us). You don't dictate every step to the employee. You tell them what to achieve and trust them to work it out.

But after reading that I kept circling something the manual doesn't say.

That autonomous "good employee", the one you don't have to explain the how to, is a senior. Someone who has spent years in the trenches, with a specific trade, judgment and scars. And the same market idealizing that employee is automating the process that manufactured them.

Quietly

SAP has put its chips on the table and there's no point arguing about it. At Sapphire this year, Christian Klein said it in plain words: the Autonomous Enterprise, where agents run the business and you focus on what really matters. They deployed 200+ agents, put up a hundred million euros for partners to build on top, and bought Prior Labs, an eighteen-month-old lab, with a commitment to invest more than a billion euros in it over four years.

SAP isn't betting on superintelligence (AGI). Quite the opposite. Klein was explicit: 80% precision may be enough for consumer AI, but not when you run the world's most critical businesses, and models shouldn't be guessing. The bet isn't an omniscient digital brain. It's a context layer — the Knowledge Graph, the data, the governance — that puts a leash on probabilistic agents which, left loose, screw up and cause real damage.

That genuinely matters so we don't drift into science fiction. What's underneath is more rudimentary and, from where I stand, more uncomfortable:

Take the accumulated knowledge of decades of consultants and processes, encapsulate it, and execute it without them.

The consultant who is both surplus and scarce

Here the numbers fight each other, and that fight is what pays off the bets in the end.

On one side, all the big consultancies following this script (as a "strange" coincidence): McKinsey announced its biggest cut since 2008 — 10% of its headcount, between 3,000 and 4,000 positions, attributing it explicitly to AI automating research and analysis work. KPMG cut around 400 advisory positions in the United States. Bain, BCG and Deloitte froze hiring. Enrique Dans summed it up without anesthesia: AI is replacing the technical expertise that for decades justified these firms as privileged intermediaries of knowledge.

Also, I read this the other day here on LinkedIn: 25% of entry-level postings in consulting and finance now require AI fluency. Two years ago it was under 5%.

On the other side, the SAP market with ECC support ending and most customers not done migrating, and senior consultant rates rising from a shortage of specialized talent...

The punchline is when somebody asks, hold on, how can it be surplus and scarce at the same time? It goes like this: scarce at the top and surplus at the bottom.

The rung we're all sawing off

Nobody is born with senior judgment, however good you are.

Judgment gets built doing the thousand boring, repetitive, guided tasks we now want to automate and are automating left and right. Seeing why a transport fails, why the IDOC didn't go through, or the class that wouldn't compile (yes, bad "old" joke, everything compiles now). Why a payment falls over because of a badly formatted field. Losing an entire afternoon on something that turns out to be an obvious configuration, and never forgetting it again because you wrote it in blood.

Expert judgment is the sediment of years of low-level work that earns you the authority to open your mouth and say something coherent.

That SAP manual I mentioned above already assumes experience as an input. You need situational judgment, dynamic reasoning, criteria. But the market that same manual describes is automating the process that produced that experience. It reserves for the human behind the keyboard the hardest thing to learn — judgment and final responsibility — and takes away exactly the low-risk practice where that judgment was forged. If you ask me right now, on the spot, I'd say it looks bad, what do you want me to tell you...

The result isn't apocalyptic unemployment (though plenty are already predicting that). For me it's something quieter: a hollow middle, which looks like this:

Expensive, scarce seniors on top, cheap agents at the bottom, and in the middle the gap where the ones who were going to replace the retiring generation used to be forming.

Bread for today, structural hunger ten years from now.

And up top, the usual paradox

Now, if the bottom rung gets sawn off, something has to happen to the people already up top. And there a strange thing shows up that anyone who's been through a restructuring knows by heart.

There's a formula that's explicit in plenty of places: more knowledge means more expensive. And when cuts come, the expensive one is the first to show up on the spreadsheet. Typical of mediocre organizations, but it's become the standard in some areas.

The trick here is guessing whether this makes seniors stay longer in the market, because there's nobody to replace them, or whether it's the push that gets them out faster, because they're the fattest line in the budget. For me it's neither, and what will happen is that the category splits down the middle.

Plenty of people are quick to call themselves senior, but there are seniors and there are seniors. Today I'll name two in particular for this case. For me there are two kinds of senior who today get paid about the same and from now on are worth very different things.

The senior whose value is accumulated speed. Does in two hours what a junior does in two days, because they've done it a thousand times. That is exactly the profile AI compresses and erases. Expensive and now substitutable, and that combination is what puts you on the list. They don't fire you for being expensive alone. They fire you for being expensive and replaceable.
The senior who holds context nobody ever wrote down. The one who knows why that field is there, why that process has a step that looks unnecessary, why the big customer invoices differently. Why the database syncs at that hour and not another. For me that person is more expensive than ever and harder to let go of than ever, because the day they leave, the reason leaves with them.

The market is already saying which of the two is scarce, and it isn't hard to guess. Consultancies cut thousands of junior positions while senior SAP consultant rates rise for lack of people. These aren't different things happening in two markets. It's the same thing happening in the same market, at two different altitudes, because the game here isn't linear or two-colored.

And this gives another reading to the famous Autonomous Enterprise we've been hearing about since the last Sapphire, because those slides, demos and emerging tech have a clear direction.

I don't think SAP designed its strategy looking at an age chart. A company that size doesn't pivot its product portfolio over that alone; Hasso's disciples are smarter than that. But demographics do explain why the message lands so well, if you want to understand it that way. The customer who can't find people is far more receptive to the "the company runs itself" show than the one with a full team. Scarcity didn't create the product. It created the demand.

What leaves me thinking, and I'll hand it to you as a question and not an accusation, is what incentive the ecosystem has to solve that scarcity when scarcity is precisely what makes the agent sellable...

What's missing from the roadmap

When a senior consultant has their work taken away, they don't just lose the salary. They lose the identity. The thing they were good at, what made them distinct in a room, what justified calling them and not someone else. Years building a specific competence, and suddenly that competence stops being what's needed from them. It's planned obsolescence building itself at scale, and it may reach quite a few people sooner rather than later.

For a lot of people of my generation in this field, the trade isn't what you do from 9 to 6. It's an important part of who you are. Part of your personality and your identity. And not because we're workaholics, but because mastery of something difficult is one of the few things that gives you back a clear identity: you're the one who knows how to solve this.

When that gets automated, the practical problem is retraining. That's why I already wrote, in one of the first articles in this space, about the danger of being a "specialist". The real problem is that you have nowhere to put that part of yourself, and that's where things can get ugly.

Corporate roadmaps solve this with one sentence: "the human focuses on what really matters". It sounds lovely and important until you ask who decides what matters, when the thing that mattered to you is no longer something you need to do. Just reading that twice short-circuits my head.

And watch out for the easy exit, because it's the one I see coming most and it sounds almost like a default:

Telling someone with years of trade that their role now is approving what the machine does is not a retraining. It's governance theater. You remove the part where the meaning lived and leave them the signature.

The honest retraining goes somewhere else: moving from operating the knowledge to being the custodian of the knowledge. The one who defines what an agent can touch and what it can't, who audits when a result smells wrong, who passes the trade to whoever comes next. That does have dignity, and on top of it it's what companies are going to need when that hollow middle starts to show.

Before you agree with me

3 things against what I just said:

This has happened before and we survived. When compilers arrived nobody cried for the people writing assembly by hand, and when Excel arrived the rooms full of people adding up columns didn't leave a crater in society and nobody is mourning them. It may be that the rung doesn't get sawn off but moves somewhere else, and that the junior five years from now comes in through a door that doesn't exist today and neither you nor I can see it.
I'm also part of the problem I'm describing (and so are you, probably). I use AI every day to do in minutes things that used to take me an afternoon. That afternoon was, for somebody, a chance to learn. I don't have a comfortable answer for that but, as they say around here: "it is what it is".
When I say human judgment and governance are needed, I'm describing work that people like me get paid to do. Keep that in mind when you read me, even if it seems a bit obvious.

What I'd do, no smoke

Three as well:

Treat a certain inefficiency as an investment. Let juniors do things the agent would do faster, precisely so they learn. It's arguably uneconomical and it's the only way to have seniors ten years from now. Whoever runs an area has to be able to defend that in a committee, and it's an uncomfortable conversation.
Change the entry skill, from executing to auditing with judgment. The new junior comes in as a kind of agent manager. But auditing well demands more judgment than executing, not less: spotting when the agent is confidently wrong, when it lacks context, when the answer is plausible and false. That judgment has to be formed beforehand, on something low-risk, or you've got a guaranteed disaster from day one.
And if you're senior, start moving your value from execution to custody now, not when they ask you to. Document what only you know. Write down why things are the way they are. Not to hand context to a machine or to the agent that will replace you, but because whoever has that knowledge organized is the one who decides what gets automated and what doesn't. It's the difference between being consulted and being replaced.

That SAP manual that made me write all of this describes the ideal employee with precision: autonomous, goal-oriented, able to adapt on its own. And without meaning to, it exposes the whole contradiction. That employee isn't born. It's made. And the same process that made it is the one we're automating.

The Autonomous Enterprise doesn't invent a new paradigm of work. It extends the old one, but without its past.

Pablo / En Producción

P.S. If you're senior and you've been feeling some of this, write to me and tell me how you're seeing it from your side. I'll use what comes in for a future edition, without names. And if you know someone sitting exactly in that spot, send it to them: this gets talked about very little and thought about a great deal in silence.

References:

SAP BTP AI Best Practices, Agentic AI section.
SAP: acquisition of Prior Labs (May 2026, closed in July).
LinkedIn Economic Graph 2026.
Forrester, "AI in Professional Services", 2025.