KPMG, Claude and Langdock: Enterprise AI is an architecture decision

Handelsblatt Disrupt asks whether the German enterprise-AI platform Langdock is underrated. KPMG is currently showing why companies need such platforms in the first place.

KPMG is rolling out Claude to more than 276,000 employees worldwide and embedding the model in its own Digital Gateway platform. A few days later, Langdock CEO Lennard Schmidt talks on the Handelsblatt podcast about a platform meant to give companies controlled access to multiple models, agents, and process automation: https://www.handelsblatt.com/audio/disrupt-podcast/disrupt-ist-langdock-die-meistunterschaetzte-ki-firma-lennard-schmidt/100227135.html

The obvious question: why would you need such platforms if KPMG brings Claude straight into its own organization?

The answer is in the architecture.

KPMG isn't simply rolling out Claude as a productivity tool for individuals. The firm can build the model deep into its own platform: domain knowledge, client data, workflows, cybersecurity, risk, and AI assurance. There, the model is only one part of the solution. The actual work sits in the control layer around it.

For large corporations that's a plausible route: their own platform, their own domain logic, their own governance, deep integration into existing work.

For mid-sized companies, law firms, and smaller advisory practices, exactly that is usually unrealistic. Not because they need it less, but because they lack the platform team, the budget, the data architecture, and the operating capacity.

That's why a new platform category becomes interesting: enterprise-AI workspaces that bundle model access, permissions, knowledge, integrations, workflows, and governance.

Langdock is just one visible example right now. A broader market of such platforms is already emerging in the German-speaking region — from horizontal AI workspaces to specialized solutions for enterprise knowledge, law firms, or legal work.

These providers often don't offer the newest model or every new vendor feature right away. But they put an enterprise layer over models from OpenAI, Anthropic, Google, Mistral, or others. That layer is often the actual thing being procured: controlled model access, permissions, integrations, data protection, admin control, standards, workflows, and cost control.

That's why the familiar "wrapper" criticism falls short. Yes, such platforms are partly wrappers. But companies need exactly this layer when AI is supposed to move out of uncontrolled individual tools and into governed work.

So the strategic question is less: which model is theoretically the best today?

For many companies it is: who builds and operates the control layer around the model?