On April 17, 129 people in McKinsey’s Berlin office voted in secret. The question was whether to appoint an election board, the step before electing a works council (Betriebsrat). Seventy-three voted no. Forty-four voted yes. Twelve abstained.

Three employees did not accept that and went to a labor court. Der Spiegel reported it, and Handelsblatt filled in the detail on October 6. The court rejected urgent relief, and the main hearing is set for July 29, 2027. According to Handelsblatt, none of the big consultancies in Germany (McKinsey, Bain, BCG, Roland Berger) has a works council.

One note on the numbers. McKinsey’s spokeswoman described a two-thirds majority. Using Handelsblatt’s figures, 73 of 117 yes-or-no votes is 62%, before abstentions. A clear majority, though not two-thirds.

What interests me is who pushed.

The plaintiffs are not on the partner track

Ben Appleton, an executive recruiter for consulting firms, made the point plainly: the initiators hold titles of Digital Expert and Director of Product, AI. Handelsblatt confirms the titles and adds that dismissal and pay proceedings were already running with some of them. They are not generalist consultants chasing partner.

Appleton reads consulting as a bargain: fast progression and high reward in exchange for long hours and wide management discretion. For the partner-track generalist it still works. For specialists it is weaker. One expert Associate Partner at a top-tier firm told him, in Appleton’s quote: “It’s very difficult to get to the next level as a technical expert.”

That is a headhunter’s account, not a study, and I do not know how any individual voted. But the tension is real at any firm that has hired AI specialists: two careers, one ladder.

Amodei’s prediction is from May 2025

A correction while we are here. The claim that Dario Amodei says AI will wipe out half of entry-level white-collar jobs is circulating again as if it were fresh. He said it to Axios on May 28, 2025: half of those jobs could go and unemployment could reach 10% to 20% within one to five years, with technology, finance, law, and consulting among the exposed sectors. That is 16 months ago. It is a prediction, and it has not expired.

What does the data say? Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at Stanford published an August 2026 update of their ADP payroll study, with data through June. Two findings sit side by side. No evidence of economy-wide job displacement. And employment of 22-25 year-olds in AI-exposed occupations is 19% below where it would be if it had tracked their less-exposed peers. It was 15% in the July 2025 data. It works through fewer hires, not layoffs.

That is not the same measure as Amodei’s, so it neither confirms nor refutes him. It is not proven cause either; the authors call it descriptive. My reading: nothing like half is visible, but the front door is closing slowly. I looked at the first cut of this data for engineering in March.

Where the two stories meet

A professional-services firm (consultancy, law firm, accounting practice, agency) runs on a pyramid. Many juniors carry the volume, some become seniors, few make partner. If agents take the work at the base, the pyramid loses its entrance. And the specialists hired to run those agents have no ladder of their own, which is exactly the complaint Appleton attaches to the Berlin case.

That leaves a question no contract answers: who trains the next senior? A senior’s judgment came from solving a hundred small cases badly and then well. If an agent solves those hundred, the junior who is left over does not learn from them. I wrote about taste debt: the bill arrives later, when nobody is left who can say the output is wrong.

What I would do in a 20 to 200 person firm

First, write the specialist ladder. A next level with a name, a way to move up without selling projects, and its own review criteria. If you do not write it, they write it themselves, sometimes in a courtroom.

Second, make junior learning a process instead of an accident. One real case a week: the junior solves it, the agent solves it, and a senior compares both against the answer they would give. Track how many cases the junior solves without the tool, even though the tool exists. It is slower and costlier than letting the agent do everything. It is also the price of having seniors in 2030.

Third, set a number for entry-level hires per year and report it like revenue. What does not show up in a partners’ meeting disappears quietly.

What we would measure first at IQ Source with a firm like that is which tasks the juniors do today, one by one, and which of them an agent could already do. That comes out of the real process map (the Process Reality Map in AI Maestro), and you need it before deciding how many juniors to hire.

If you run a services firm and that task list does not exist yet, map those tasks with AI Maestro