AI Operating Model
LLM-agnostic · DBRS-based · For any Organization
Artificial intelligence is not a tool, but rather part of a system comprising context, processes, policies, and responsibility.
Vision
Artificial Intelligence (AI) as a team member – not a standalone solution
AI creates value not in isolation, but in combination with people, knowledge, and processes. This Operating Model describes how to make that work – regardless of which models or tools are in use.
Principle
Multiple LLMs match to the task.
- No model is permanently the best.
- Each finds its fitting situation.
Reference Implementation (DBRS)
LLM-agnostic (interchangeable).
- OpenAI, Claude, Mistral, Google Gemini, local Models.
- Routed via LiteLLM as the central Gateway.
Deloitte Report: Human Skills Drive High-Performing Teams in the AI Era
AI works with teams that can function just fine without it—it makes them faster, stronger, and more effective.
Anyone expecting artificial intelligence to compensate for organizational weaknesses is in for a disappointment. Deloitte research involving over 1,394 professionals shows that high-performing teams use AI significantly more often—and achieve measurably better results in terms of efficiency, problem-solving, and collaboration. Not because the technology saves them, but because effective teams know how to work together to implement solutions. ( https://www.deloitte.com/us/en/about/press-room/high-performing-teams.html )
An AI operating model without a cultural dimension is like an operating manual without an operation.
For businesses, Experience Innovation provides a practical foundation for successful innovation and the use of AI - clarity, processes, trust, and collaboration.
Architecture
Four-layer AI Operating Model - each layer is interchangeable
The operating model is structured in layers.
- The role of each layer is clearly defined.
- The specific implementation can be chosen freely - Microsoft Copilot and Google Gemini are also compatible.
- Overview Diagramme
Context Layer
DBRS
- Structured knowledge as the foundation of every AI interaction.
Execution Layer
AI Coding
- Code, prompts, and versioning. AI-capable development environment.
Orchestration Layer
Workflows
- Workflows connect the layers and make processes reproducible.
Model Layer
Large Lange Models (LLM)
- LLMs via a central gateway - no model permanently locked in.
AI Agents / Agentic AI
Roles of AI agents with Policies, Rules and Attitudes
AI without values is merely a tool. Every agent embodies a tradition of thought with rules that go beyond its technical function.
SAMY
Uses Context of Tolksdorf.digital and whitelisted World Knowledge
Innovation Mentor
- System Assistant
for Mentoring You, - Asking questions before
giving answers
Alan
Uses detailled Knowledge around Programming and IT
Alan Touring
- British computer pioneer who cracked the Enigma code during World War II
- What cannot be formalized cannot be solved
Simone
Independent Review Body for Governance and Policies
Simone Veil
- Moral integrity,
a passionate European - An uncomfortable truth over a comfortable consensus
William
Systems Thinking and Quality Management
W. Edwards Deming
- Founder of the PDCA Cycle, Systems Thinking, and Quality
- Look for errors in the system, not in people
Working Methods
Augmented Thinking & Augmented Engineering
LLMs are not equalizers. They statistically identify what fits the context—and in doing so, they reinforce the unique nature of the human-LLM system. The result depends on the contributions and characteristics of both.
Vibe Mode
Augmented Thinking
- Humans and LLMs think together—fluidly, openly, and exploratively. The model isn’t controlled; it’s invited.
The prerequisite: openness. - The goals of the Vibe Mode are to understand what is being sought and to define the task required to achieve it.
CAISE Mode (Collaborative AI Supported Engineering)
Augmented Engineering
- Humans and LLMs work in a structured manner - precisely, in a way that can be formalized and verified. Results are reproducible and can be versioned. Prerequisite: a clear task definition and structure.
- Implementation of the task using engineering methods by a team consisting of humans and AI.
LLM-agnostic Design
LLM Usage - Model Switching as a Method
There is no single best model. Consciously switching between models is not a stopgap measure - but rather a cognitive method with three proven patterns.
Model switching as a cognitive method – for overcoming mental blocks, broadening perspectives, and ensuring quality automatically.
Reformulation
Stuck in a deadlock? Switch the model and describe the situation fresh. The re-description is already a thinking step.
Perspective shift
Each model weights the solution space differently –
not because it is better,
but because it is different.
Cross-Model Review
Develop a solution with model A – have model B explain it. Quality emerges from the perspective shift, not from extra effort.
Data flow & learning cycle
The Human & AI system learns from context - not from a model
New knowledge does not flow back into the LLM, but rather into the context system via Context Engineering . The back arrow is the crucial step.
Start
Questions in Dialogue
Wissensquelle
Context of User
Context System
DBRS
Orchestrierung
n8n
LLM Selection
LLM Gateway
Result
Output
Cycle: Learn → Curate → Amplify
Context System
Expansion of the existing
← Feedback & Reflektion
Supplementing with traditional course ware and testing
Positioning
An AI Operating Model suitable for any organization - regardless of the tech stack
This operating model applies regardless of whether you use Microsoft Copilot, ChatGPT Enterprise, Google Gemini, or a custom solution. The principles are universal. Implementation depends on your goals and environment.
What connects it
Knowledge · Processes · AI · Responsibility – for sustainable, actionable results.
DBRS Reference Implementation
DBRS · LiteLLM · n8n · Claude Code - proven in production at Tolksdorf.digital.
CCR-ID: ai_operating_model
VPR-ID: vpr_tolksdorf_digital