Our AI manifesto

We sat down to discuss a strategy and agree on how Glashaus handles AI. Let us give you the tour.

Five concerns we take seriously

Data protection

We don't handle highly sensitive data, but we protect the anonymity of everything clients and research partners share with us. As a German company, we answer to the GDPR and the EU AI Act.

Global resources

AI data centres already use more energy than entire countries, and they cool servers with water that people and ecosystems need. We take responsibility for our share of that footprint.

Data protection

We don't handle highly sensitive data, but we protect the anonymity of everything clients and research partners share with us. As a German company, we answer to the GDPR and the EU AI Act.

Global resources

AI data centres already use more energy than entire countries, and they cool servers with water that people and ecosystems need. We take responsibility for our share of that footprint.

Our core skills

Brains that always get solutions served stop developing their own. Structured, context-informed problem solving is where we excel and why clients come back. We keep that muscle in training.

Quality and bias

Our voice is part of why clients work with us. Insights shaped by an LLM turn generic and flat. Commercial models also carry the biases of their training data, and we refuse to reproduce them.

Dependency

AI develops faster than anyone can follow. Workflows built around one model drift when the model does. We don't want to depend on one LLM, one provider or one tech giant.

Our core skills

Brains that always get solutions served stop developing their own. Structured, context-informed problem solving is where we excel and why clients come back. We keep that muscle in training.

Quality and bias

Our voice is part of why clients work with us. Insights shaped by an LLM turn generic and flat. Commercial models also carry the biases of their training data, and we refuse to reproduce them.

Dependency

AI develops faster than anyone can follow. Workflows built around one model drift when the model does. We don't want to depend on one LLM, one provider or one tech giant.

So why use it at all?

A capable assistant

LLMs analyse data, spot patterns and summarise text well, and they keep improving. With good prompts and clear guardrails, they take repetitive work off our desks.

Speed

AI doesn't sleep or take days off. We deliver first insights faster and react quickly when a project changes course.

Collaboration

Working with AI pushed us to store knowledge in accessible ways. Anyone on the team can pick up work where someone else left off.

Creativity

Glashaus is a space to experiment. AI is our sparring partner for bold ideas, off-track thinking and early concepts.

Seven principles, in detail

How we live our values in the face of these concerns.

1

Differentiated

We decide deliberately where AI leads, where it works alongside us, where it only supports and where it has no business at all. AI leads on repetitive, scalable work like scouting and proofreading. It stays out where trust, presence or judgment are the deliverable: client relationships, fieldwork, workshop facilitation. That keeps our own brains busy, too.

2

Human in the loop

Everything AI has touched gets reviewed by us before it carries our name. Charlotte or Aline edits and approves every AI-assisted report, proposal and analysis. We stand behind what we publish, individually and as a team.

3

Transparent

We're open about where and how we use AI, internally and externally. For published work, we note which parts AI assisted and who reviewed them. The butterflies on our homepage? Generated with AI, and labelled as such.

4

Independent

We don't tie ourselves to a single model or provider. We keep our tools diverse, never make all our data accessible to one system, and audit our setup twice a year to avoid lock-in.

5

Resource-conscious

We pick the right model for the task, not the biggest one. Small models polish emails and summarise single documents. Larger ones take on research synthesis and complex systems work. We reuse results and prompts through reference files, a prompt library and batch processing.

6

Self-directed

We shape AI around our way of working, not the other way round: through our own workflows, our quality standards and, where it makes sense, our own agents. That's how our voice stays ours.

7

Legally sound

We work GDPR and EU AI Act compliant. Client data, interview material and documents under NDA only enter an AI tool once we've checked its data-handling terms. AI-generated content gets a label wherever the law requires one.

In practice

What we know

AI misstates facts, reproduces bias and sounds confident while being wrong. We check everything factual, numerical or attributable before we use it. Neither of us treats AI output as a source of truth.

What we use

AI assistants built on large language models, for research, drafting and synthesis. Image generators for visuals and early concepts. AI features inside our design and workspace tools. We treat every output as a fast first draft, never a finished product.

What we don't do

We don't put confidential client data, personal data or NDA material into a tool before checking its terms. We don't send unreviewed AI output to clients. And we don't let AI make the calls that belong to us: strategic recommendations, evaluation findings, client-facing conclusions.

Why this exists

This manifesto is our AI literacy practice under Article 4 of the EU AI Act. We revisit it whenever our tools, workflows or client requirements change. Curious how AI would figure in your project?

Talk to us

Discover how your innovations

can thrive.

LinkedIn
hi@weareglashaus.de