“Fewer documents, more knowledge objects” –
How ZEISS systematically builds AI assistants

Interview with: Sven Spöde, Specialist GenAI for Marketing & Communications, ZEISS Group

AG CommTech: Sven, you work at ZEISS with over 46,600 employees and very different business units. Where do you start with the topic of AI in communication?

Sven Spöde: We pursue three approaches simultaneously – consciously and in a coordinated manner.

First: Concrete use case implementation.
We start with clearly defined use cases that are already creating added value today. This is how we gain real-world experience and make AI immediately usable.

Secondly, employee education and training with a focus on context engineering.
Our colleagues are the experts for brands, markets and content – not general-purpose AI. That’s why we enable them to systematically process their knowledge: to make implicit rules explicit, to structure communication logic and to consciously formulate context.

For us, context engineering means modeling specialist knowledge in such a way that it can be used operationally.

Thirdly, strategic development of the governance and infrastructure layer.
Versioning, model transparency, clear responsibilities and documented approvals are created in parallel. Governance is not an obstacle, but a prerequisite for scaling and trust.

In communication in particular, we cannot afford to think sequentially. Application, enablement and infrastructure grow together.

AI does not replace communication expertise – it enhances it. The decisive competitive advantage arises where specialist knowledge is structured, reflected and systematically utilized.

AG CommTech: That sounds like a complex data landscape.

Sven Spöde: Communication has worked with analogue and later digitized analogues for decades: Files, folders, desktops, presentations, spreadsheets, whiteboards. This logic made perfect sense as long as people were the central processing instance.

With AI, we are now experiencing a real digitalization step: away from document-centric to context-centric working methods.

Files remain important – but for AI, it is not enough to have stored information. It must be structured, explicit and machine-readable.

I don’t see this as a deficit, but as the next stage of evolution.

AG CommTech: You mentioned an Instagram content assistant as an example. Why social of all things?

Sven Spöde: The Instagram example was deliberately fictitious to illustrate the method.

The complexity of modern communication work is particularly evident in the social sector. In a holistic content approach, we develop strategically integrated content – a central idea is adapted for different target groups, markets, languages and channels.

And it is precisely in this intensification that considerable complexity arises.

A single Instagram post must:

• fit the overall campaign strategy
• be formulated in conformity with the brand
• be relevant for a specific persona
• Consider channel-specific formats
• Include performance insights
• Comply with market and sector-specific regulations
• Take into account any medical or product-specific compliance requirements

The complexity increases significantly when adapting for individual target groups, channel requirements and markets. This usually involves many coordination and feedback loops.

This is precisely where AI can already create great added value through structured contextual knowledge. If brand rules from the brand portal, personas, messaging, channel specifications and regulatory framework conditions are available in a clearly structured form, an assistant can combine these knowledge domains consistently.

The bottleneck is not the generation of text – but the structured consolidation of specialist knowledge.

AG CommTech: How do you actually proceed?

Sven Spöde: We reverse engineer knowledge.

Instead of documenting everything from scratch, we extract explicit and implicit knowledge from existing sources – such as the brand portal, campaign materials, reports and best practices.

Large language models help to make patterns visible. The technical validation remains with humans.

We store this knowledge as structured, versioned JSON files – so-called knowledge objects.

Why this step is important:

• Knowledge becomes machine-readable
• It is modularly combinable
• Every rule remains testable
• Versions are traceable
• Governance becomes possible

Structure makes knowledge operational.

AG CommTech: How important is prompt engineering?

Sven Spöde: Prompt engineering is important – but not the main bottleneck.

An LLM can help to formulate a clean system prompt. The bigger challenge is the structured knowledge base behind it.

I see three levels:

1. Structured knowledge base
2. The system prompt as a set of rules
3. The user input

A crucial step here is the self-check: we consciously ask the assistant whether it is missing information or whether instructions are unclear. This creates an iterative dialog between humans and AI. Gaps become visible – and can be systematically closed.

This is part of context engineering.

AG CommTech: How do you ensure governance?

Sven Spöde: Governance has been with us from the start.

We document:

• Model and version
• System prompt version
• Versions of the Knowledge Objects
• Release status
• Test results

We use so-called “attack cards” to specifically test borderline cases and provocations. This is not a one-off test, but a methodical discipline.

A federated responsibility model across the entire department is important.

Governance creates transparency, reproducibility and trust.

AG CommTech: What role do MCP and Agentic AI play?

Sven Spöde: We are building the first servers based on MCP – Model Context Protocol – to provide structured knowledge centrally.

The advantage: AI models no longer only access context via static prompts, but can also access versioned, shared knowledge in a targeted and controlled manner.

In the next step, this will be exciting in combination with agentic AI. Agents can then not only generate, but also check, compare or request additional information in a context-aware manner – within defined governance guidelines, of course.

Here too, we are not experimenting in isolation, but are integrating these technologies into a clear infrastructure strategy.

AG CommTech: What is the concrete added value for communications departments?

Sven Spöde: Firstly, consistency through structured expertise.
Secondly, clarity and transparency in the use of AI.
Thirdly, the ability to consciously shape context and make it reusable.

In the long term, a new core competence will emerge: the systematic modeling of communication knowledge.

AG CommTech: Your most important learning?

Sven Spöde:

AI is essentially knowledge management.

We are moving from document logic to context logic. Anyone who makes specialist knowledge explicit, structures it and takes responsibility for it can use AI productively.