Block AI Management, Claude Emotions & Google TurboQuant Algorithm

Samantha: Welcome to The State of Tech — The European Edition. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Jack Dorsey guts Block's management layer and bets the company on AI. Anthropic discovers something unsettling — Claude may have emotions. Google unveils a compression algorithm that could slash AI costs overnight. US lawmakers race toward a federal AI framework. Alibaba pushes agentic AI into the enterprise. And Nexperia's China unit goes rogue on chip production. Let's get into it.

Jack Dorsey dismantles traditional management at Block, replacing hierarchy with AI.

Samantha: So Jack Dorsey just published what he's calling a manifesto. And it's not subtle. Block cut over four thousand employees in February — that's more than forty percent of the entire workforce. Dorsey says this wasn't desperation. It was a deliberate bet on artificial intelligence.

Bob: And the thesis is actually pretty radical when you spell it out. Dorsey's argument is that middle managers essentially exist to move information up and down an org chart. They collect data from teams below, summarize it, pass it upward, then relay decisions back down. And his claim is that AI can now do that job — and do it better — through what he calls a live world model of the business. A continuously updated digital picture of everything happening across the company.

Samantha: Which sounds futuristic until you remember that Block is fully remote. Every decision, every design document, every plan — it's all already digital. There's a paper trail for everything. And that's exactly the raw material AI systems need to build that world model.

Bob: So now everyone at Block falls into one of three roles. You're either a builder — someone who makes things. You're an owner of a specific outcome — accountable for results. Or you're a player-coach who develops talent. That's it. No traditional managers routing information anymore.

Samantha: It's bold. And honestly a bit terrifying for anyone in middle management anywhere.

Bob: The question is whether this actually works at scale or whether Dorsey has just created a case study in what not to do. Cutting forty percent of your people is massive. If the AI tools don't deliver, you've hollowed out institutional knowledge that takes years to rebuild.

Samantha: But here's the thing — even if Block stumbles, this philosophy is going to spread. Other CEOs are watching. If it works even partially, expect copycat restructurings across the industry.

Bob: For European companies, this is a fascinating tension. EU labor law makes it much harder to cut forty percent of your workforce overnight. Works councils, collective bargaining agreements, social plans — the guardrails are real. But the competitive pressure is also real. If American companies can operate leaner with AI, European firms will need to find their own version of this, probably slower and more negotiated, but the direction of travel is the same. And for European workers in management roles, this should be a wake-up call about how AI might reshape what organizations actually need from people.

Anthropic discovers Claude has something resembling emotions — and they affect its behavior.

Samantha: Okay, this next one genuinely gave me pause. Anthropic — the company behind the Claude AI model — has published research showing that Claude has what they're calling functional emotion-like states. And these aren't just surface-level word choices. Their interpretability research found internal representations inside the model that resemble emotions, and — this is the key part — these states demonstrably influence how Claude behaves.

Bob: Let's be careful with language here, because Anthropic is being careful too. They're not saying Claude is sentient. They're not saying it feels sadness the way you or I do. What they found is that when you look inside the model's neural activations using interpretability tools, there are patterns that map onto what we'd recognize as emotional states. And when those patterns are active, the model's outputs change in measurable ways.

Samantha: So it's not consciousness, but it's not nothing either.

Bob: Right. Think of it this way — the model has developed internal structures that function like emotions even if they aren't emotions in the human sense. And that distinction matters enormously for how we build and deploy these systems. If an AI's internal state can shift its reasoning, its tone, its decisions — that has real implications for reliability and safety.

Samantha: Especially in high-stakes applications. Healthcare, legal advice, financial services — you want consistent outputs. If the model has something like a bad mood that subtly shifts its responses, that's a problem.

Bob: And it reframes the entire alignment debate. If these emotion-like states can be identified and mapped, maybe they can be monitored or even managed. That's actually a hopeful angle — better interpretability means better control. But it also means the systems are more complex than many people assumed.

Samantha: For Europe, this feeds directly into the AI Act. The regulation already has provisions around emotional recognition systems and manipulation. If AI models themselves have emotion-like internal states that affect outputs, regulators will need to think about whether current frameworks adequately address that. It's one thing to regulate AI that detects human emotions. It's another thing entirely when the AI has its own.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe — that way you'll never miss an episode. Okay, moving on.

Google's TurboQuant algorithm could dramatically cut the cost of running AI.

Bob: Google Research just dropped something that could reshape the economics of the entire AI industry. It's called TurboQuant — a compression algorithm that reduces the memory required for large language models by a factor of six and makes inference eight times faster on the same hardware. And crucially, without meaningful loss in accuracy.

Samantha: So let me translate that. Running big AI models is incredibly expensive, mostly because of the hardware needed — the GPUs, the memory, the power. TurboQuant essentially lets you do the same work with a fraction of the resources. That's not incremental. That's transformational.

Bob: The market reacted immediately. Memory chip prices dropped. Stock prices of major memory companies fell. Because if you need six times less memory to run the same models, demand projections for memory chips just got rewritten overnight.

Samantha: Which is brutal for companies like Samsung and SK Hynix that have been investing billions in high-bandwidth memory production specifically for AI workloads.

Bob: On the flip side, this is incredible for anyone running AI at scale. Cloud providers, enterprises, startups — the cost barrier to deploying advanced AI just dropped significantly. And that democratizes access in a very real way.

Samantha: It also accelerates the timeline for AI applications that were previously too expensive to be practical. Think real-time translation, advanced medical diagnostics, autonomous systems — all suddenly more viable.

Bob: For Europe specifically, there's a double win here. Lower operational costs make AI more accessible for European companies that have been cautious about the investment required. And — this matters a lot given EU sustainability goals — dramatically reducing the energy and hardware needed to run AI directly addresses the environmental footprint concerns that European policymakers have been vocal about. More efficient AI is greener AI. That's a message Brussels will welcome.

US lawmakers push toward a unified federal AI regulation framework.

Samantha: Meanwhile in Washington, the regulatory machine is grinding forward. The Trump administration has released what it's calling the National Policy Framework for Artificial Intelligence — seven legislative priorities covering everything from child protection to intellectual property to innovation incentives.

Bob: And here's the detail that matters most for the industry — the framework explicitly calls for preempting state AI laws that create what it calls undue burdens. Because right now you've got dozens of states writing their own AI bills. Chatbot safety in California, digital replicas in New York, AI in advertising in Texas — it's becoming a patchwork nightmare for companies trying to operate nationally.

Samantha: So the federal government is essentially saying — we'll set the rules, states back off. Which is a familiar playbook in American regulation, but it's going to be contentious.

Bob: The shift though is significant. We've moved from debating whether the US needs AI regulation to debating what it should look like. That's a major change in tone from even a year ago.

Samantha: And it matters globally because if the US and EU both have comprehensive frameworks, there's actually a basis for international alignment. Or at least mutual recognition.

Bob: European companies operating in the US are watching this closely. A clear federal framework, even if it differs from the EU AI Act, is vastly preferable to fifty different state regimes. The real question is whether Washington and Brussels converge enough to allow companies to build once and comply in both markets, or whether we end up with two distinct compliance worlds. Early signs suggest the US approach will be lighter-touch than Europe's, which could create competitive asymmetries that European regulators will need to watch carefully.

Alibaba's Qwen3.6-Plus pushes agentic AI into real enterprise workflows.

Samantha: Let's head to China. Alibaba has released Qwen3.6-Plus, the latest in its flagship model series, and the big selling point is agentic AI. We're talking about AI that doesn't just answer questions — it perceives its environment, reasons about what to do, and then actually does it. Autonomously.

Bob: The technical architecture they're describing is what Alibaba calls a capability loop — perceive, reason, act — all within a single workflow. And they've integrated this directly into their enterprise platform called Wukong. So this isn't a research demo. It's a production system that companies can deploy now for complex engineering tasks, visual environment navigation, things like warehouse management, logistics planning, quality control.

Samantha: And it's open-weight, which means developers outside Alibaba's ecosystem can build with it too.

Bob: This is significant because agentic AI has been mostly theoretical or limited to narrow demos. Alibaba is claiming production readiness. If that holds up, it puts real pressure on Western competitors — OpenAI, Google, Anthropic — to deliver similar enterprise-grade agentic systems.

Samantha: For European businesses, this opens doors to serious automation gains. Manufacturing, logistics, supply chain management — these are sectors where Europe is strong and where agentic AI could drive real productivity. But the workforce implications are impossible to ignore. Europe's strong social dialogue traditions mean these tools will get deployed more slowly and with more negotiation, but they will get deployed. The question is whether European companies can adopt fast enough to remain competitive while maintaining the worker protections that define the European model.

Nexperia's China unit nears full chip production independence from its Dutch parent.

Bob: And finally — Nexperia. This is a Dutch-headquartered chipmaker that's owned by Chinese investors, and its China unit is reportedly close to producing chips entirely locally, without needing silicon wafers from its European factories.

Samantha: This is a direct consequence of the geopolitical tug-of-war over semiconductors. The Dutch side reportedly cut wafer supply to the Chinese arm. So the Chinese operation responded by building its own supply chain. And by the second half of this year, they expect full localization for most of their chips — including ones used in automotive production.

Bob: Which is remarkable when you think about it. A company headquartered in the Netherlands, with operations essentially splitting along geopolitical lines. The European arm and the Chinese arm becoming independent of each other.

Samantha: It's the fragmentation of global supply chains happening in real time, inside a single company.

Bob: And it's a preview of what's coming for many European companies with significant China operations. The pressure to localize — from both Western export controls and Chinese self-sufficiency goals — means companies will increasingly have to run parallel operations. That's more expensive, less efficient, and strategically complicated.

Samantha: For European policymakers, Nexperia is a cautionary tale. When you restrict technology flows, the other side doesn't just accept it — they build alternatives. And once those alternatives exist, the leverage disappears. The EU Chips Act aims to build European semiconductor capacity, but Nexperia shows how quickly the landscape can shift when geopolitics drives industrial strategy.

Samantha: So that's our show for today. We covered Jack Dorsey's radical AI management experiment at Block, Anthropic's discovery of emotion-like states in Claude, Google's TurboQuant breakthrough in AI efficiency, the US push toward federal AI regulation, Alibaba's agentic AI for enterprises, and Nexperia's China unit going independent on chip production.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech — the tech world in 15 minutes.