"AI without context is stupid" – Why the smartest models in the world fail in everyday business

Germany's AI Transformation | Part 1 of 3

GPT-5 knows what a supply chain is.

It doesn't know that your supplier in Bratislava requires a lead time of 14 business days for orders exceeding €50,000. It doesn't know that your internal approval matrix requires both the Head of Procurement and the CFO above that threshold. And it doesn't know that the current EU Supply Chain Directive triggers a specific documentation obligation for that very supplier.

Ask it anyway, and you'll get an answer. One that sounds good, is professionally worded, and is irrelevant at best – factually wrong at worst.

Welcome to the reality of AI in everyday business.

The Race Everyone Was Watching – and the One That Actually Matters

The global AI debate in recent years has been dominated by a single question: Who builds the most powerful models? That question has been answered. A handful of American companies won. The race for foundation models is, for all practical purposes, over.

But with that answer comes a far more uncomfortable realization: A Large Language Model that knows everything and understands nothing is virtually worthless in everyday business.

The knowledge that companies actually need isn't stored in the models. It resides in ERP systems, CRM databases, internal wikis, compliance documentation – and in the minds of the most experienced employees. LLMs have no structural access to this knowledge. They were trained on publicly available data, not on the proprietary, highly specific knowledge that companies have built up over decades.

AI without this context is stupid. Not malicious, not dangerous – simply useless for the decisions that need to be made every day.

When AI Starts Guessing: Hallucinations and "AI Sloppiness"

It gets even worse when the model doesn't acknowledge its knowledge gaps but starts to hallucinate – generating answers that sound plausible but are factually wrong.

The consequences are insidious: Employees who have received a hallucinated answer once lose trust in the system. Or – and this is the more dangerous scenario – they continue using it without verifying the answers. Experts call the result "workslop" or "AI sloppiness": pseudo-work that simulates quality but doesn't deliver it.

For companies, this is not a theoretical risk. It is the reason why many AI pilot projects are quietly shelved after the initial euphoria – not because the technology is bad, but because without the right context, it doesn't deliver what it promises.

What Companies Actually Need: Not One Model – Many

The implication is profound: A single, generic model cannot meet the requirements of an enterprise. What organizations actually need are multiple domain-specific models, each tailored to the language, data, and regulatory framework of a specific functional area:

• Procurement needs a model that knows framework agreements, pricing histories, and supply chain due diligence legislation.

• Sales needs one that understands CRM data, customer history, and competitive intelligence.

• Manufacturing needs one that masters machine data, quality protocols, and production tolerances.

• Logistics, maintenance, service – each area has its own data, its own rules, its own risks.

And each of these models must additionally be enriched with external knowledge: industry standards, regulatory requirements such as the EU AI Act or the GDPR, current market data. This knowledge is not static – it changes when laws change, when new standards come into effect, when markets shift.

The Real Question

When a single model isn't enough. When companies need domain-specific models built on proprietary knowledge. When that knowledge is distributed across dozens of systems, constantly evolving, and must be used under strict regulatory conditions…

…Who orchestrates all of this?

Who ensures that the right model receives the right data, under the right conditions, for the right user – and that every decision remains auditable?

This question leads to a new software category that analysts consider larger than CRM and ERP combined. And it leads to a historic opportunity for Germany.

But more on that in Part 2: The Invisible Layer – What Companies Really Need Between AI and Value Creation.

This article series is based on the position paper "Enterprise AI Management & Orchestration" by Karl-Heinz Land, CEO & Founder of neuland.ai AG.

Image generated using the neuland.ai HUB.