Jonas Andrulis: "Digitize the state! That’s the foundation we all stand on"

The co-founder of Aleph Alpha argues the real risk of AI isn’t job loss – it’s speed, concentration of power, and whether Europe chooses to build or just buy

Jonas Andrulis is a co-founder of Aleph Alpha, a German AI company. It produces a sovereign, end-to-end generative AI platform designed for enterprises and governments to create, deploy, and manage custom AI solutions within their own environments. Aleph Alpha has a strong focus on European values such as transparency, data protection, and explainability. As one of the leading European players in the field, the company is seen as strategically important for the continent’s digital sovereignty and innovation capacity. It is financially backed by both private investors and public funding, underscoring its relevance on the European tech landscape.

Jonas, let’s start high-level: What is AI doing to the labor market right now?

Think of AI as another industrial revolution. Past revolutions re-organized physical labor. This one is reordering knowledge work – information-based value chains. So when we talk about impact, we’re mostly talking about white-collar roles.

Which sectors will feel it first, say over the next two or three years?

In truth, every knowledge sector. But the capabilities are uneven. Today’s general chatbots do some things well – drafting, editing, summarizing – especially when deep domain expertise isn’t required. That’s why you see immediate change at entry level roles. The classic first rung – junior staff doing basic research, writing first passes that seniors review – is in the middle of being completely transformed. Law is a good example: the traditional early-career diet of “read, search, formulate” is simply less necessary in that form. Training and career ladders will have to adjust.

There’s an ongoing debate: Will AI erase jobs, or just tasks?

“AI will eliminate jobs” isn’t quite right. Jobs evolve. I still remember calling libraries to check if a book was in stock, then pressing pages onto a copier. Nobody misses that. The pattern is familiar from the PC and the internet: activities change, professions reshape. The real risk isn’t that this happens; it’s how fast – and whether companies and societies can adapt at that speed. Add one more risk: power and value further concentrating in a few trillion-dollar companies in the U.S. and China. That concentration, not technological progress per se, should worry us.

If the change is inevitable, what should companies actually do?

Stop waiting. Lead the transformation. Re-invent your own operating model so you’re still strong five years out. The imperative isn’t headcount defense; demographics alone mean we’re short on qualified people. The imperative is innovation, productivity and resilience.

There’s no shortage of pilots. Why do so many stall?

Because a “general-purpose chatbot” rarely solves the hard parts. If your benchmark is “writes a cute poem” or “drafts a simple email,” fine. But for our customers – in government, security, finance, industrial manufacturing – the hard problems are specialized, safety-critical, and context-heavy. Many firms tried to hook a big-tech chatbot to internal data and call it an enterprise solution. What you don’t get with this appoach is high-precision, domain-specific judgment – especially in areas rarely or never covered in publicly available data on the internet. Even with fine-tuning, these systems tend to sound confident when they should admit “I don’t know.” That’s not acceptable where the stakes are high.

So what are the use cases that do work?

Industrial R&D and production are compelling – and strategically vital for Europe. Picture the classic V-model: from requirements and design down the left side, then verification and validation up the right. Based on our end-to-end AI stack PhariaAI, we’ve built solutions across those steps that let experts query complex documentation, models, and test evidence in ways that compress time to insight and improve decision quality. Yes, sometimes the interface is a chat, but often it’s richer –experts can handle complex software and uncertainty. We’re in production in select environments and broader evaluation elsewhere. Think of it as the early innings of a long transformation of core processes.

Tech isn’t the only barrier. Big organizations have culture and silos. What’s the biggest hurdle to scaling AI: technology or management?

Both. On the tech side, getting real domain knowledge into models is non-trivial. That’s why we built a tokenizer-free architecture – to better ingest information that’s truly out-of-distribution relative to web-scale training sets. On the org side, it’s change management 101 – but everywhere, all at once: collaboration models shift, roles evolve, people ask where they fit. Skepticism and inertia are human; they have to be managed, not wished away.

Europe brands itself as an “AI continent.” There’s the Data Act, talk of pan-European data spaces, and so on. Are we positioned to compete?

It’s telling that when we say “well-positioned,” the first thing we list is regulation. We have plenty of it – cookie banners, data protection regimes that almost no one fully satisfies. The ideas – data spaces, data sharing – are good. Execution is spotty. Before we open-sourced what is likely the best German dataset of its kind, it wasn’t easy to find large, high-quality datasets. There are bright spots: we’re partnering with Data Hub Europe – announced at the Digital Summit by Deutsche Bahn and Schwarz Group – to enable sovereign, secure sharing of critical data. That’s the kind of infrastructure that matters.

You keep using the word “sovereignty.” What do you mean by it?

The ability to take responsibility – because you can understand and act. You need knowledge, and transparency, the ability to stay in control with your own teams and infrastructure, and the capacity to capture a fair share of the value created. The hyperscaler model, if you adopt it wholesale, turns you into a very sophisticated user of technology built and governed elsewhere. That’s not sovereignty. And this isn’t just an economic issue. AI systems will mediate how the next generations think, talk, and relate—companionship is already the top consumer use case. If a handful of players define “right answers,” they shape public discourse. Liberal democracies can’t afford to take their hands off the wheel.

What market structure best manages AI risk – heavy regulation that tends to favor incumbents, or a looser regime with more open-source that’s harder to control?

Beware of regulatory capture. The more red tape, the more entrenched players benefit. Europe won’t catch up by making entry harder. And we should be honest about costs. Rules aren’t free; GDPR-class compliance just like every other new regulatory action burns giant sums and diverts scarce creative energy – from building products and new business models to paperwork and controls. The question isn’t “protection or no protection”; it’s the right balance that protects people and enables innovation.

Speaking of giant sums: Europe wants to fund more data centers. Smart investment – or buying GPUs from NVIDIA to park them in the EU while calling it “sovereignty”?

Having infrastructure isn’t bad. But today, GPU availability isn’t the bottleneck. If you want GPUs – including in Europe – you can get them. Prices for H100 capacity have fallen sharply in the last six months – roughly from €2.50 to €1.50 per GPU-hour. Are we about to spend a lot of money on hardware that depreciates fast, to alleviate a shortage that’s easing? Buying a ton of hardware that may be obsolete in two or three years alone won’t buy “tech sovereignty.” Meanwhile, the real returns come from everything that turns compute into sustainable national capability: digitized public services, industrial processes, security workflows. Modernize a permitting process with AI and you don’t get a one-off benefit—you get compounding returns for decades. Buying servers doesn’t equal strategy. Building value chains does.

Let’s land this with one concrete ask. Germany has a Digital Minister for the first time. If you could put one priority on Karsten Wildberger’s desk, what would it be?

Digitize the state – administration and security services with a sovereign strategy. That’s the foundation we all stand on. If public services stall, frustration rises, and that’s politically toxic. AI can help here immediately. We’re involved in several projects, and many strong European vendors can deliver. From what I see, the ministry has this near the top of the agenda. It should stay there.

Key Takeaways

Jobs morph, don’t vanish – but velocity matters. The disruption is less “no jobs” and more “new jobs, new ladders”, especially as entry-level tasks get automated. The danger is failing to adapt fast enough and ceding value to a few platforms. 

Generic chatbots won’t run the enterprise. Safety-critical, high-context work needs domain-aware systems and workflows beyond a text box. Industrial V-model use cases are promising early wins. 

Sovereignty is capability, not a data-center postcode. Europe should invest less in raw hardware bragging rights, more in data sharing, public-service digitization, and industrial AI that compounds over time.

---

Season 1: AI and the Labor Market | Episode 1: The Future of Work, In Progress | Episode 2: Carl Benedikt Frey: “Professionals are not prepared for the coming changes”