The Core-and-Floors Model: an enterprise AI reference architecture framework.
Production AI is a building, not an app. Five floors carry each other from data to experience, and a governance and security core runs through all of them. This is how we draw it, assess it and build it.
The framework at a glance
Each floor depends on the one beneath it. The core is not a top layer to add at the end; it is designed into every floor from the first diagram.
- ExperienceCopilots and assistantsCustomer channelsAI inside core appsWhere people meet the system. Designed around the job, with clear hand-offs to humans.
- Agents and orchestrationAgent runtimeTool and MCP gatewayHuman approval stepsAgents that act through governed tools, with durable workflows and checkpoints you control.
- Models and gatewayLLM gateway and routingGuardrailsEvaluation harnessOne front door to every model, with policy, cost attribution, fallbacks and evals built in.
- Knowledge and retrievalIngestion pipelinesHybrid search indexPermission-aware retrievalAnswers grounded in your content, filtered by what each user is allowed to see.
- Data foundationLakehouse or warehouseData contractsLineage and qualityTrustworthy, governed data products that AI workloads can depend on.
- Identity
- Policy
- Audit trail
- Observability
- Cost control
Floor 5
Experience
Where people meet the system: assistants, copilots inside existing tools, customer channels and AI features embedded in core applications.
How we design this floorCore components
- Assistant and copilot interfaces
- Embedded AI features in line-of-business apps
- Customer-facing channels with clear hand-off to people
Decisions to make
- Which jobs get an assistant versus an embedded feature
- Where a human must confirm before anything happens
- How users report a wrong answer
Warning sign: A chatbot bolted onto the intranet that nobody owns and nobody measures.
Floor 4
Agents and orchestration
The layer that lets AI act, not just answer: agents, workflows and the tools they are allowed to call.
How we design this floorCore components
- Agent runtime with durable, resumable workflows
- Tool and MCP gateway with scoped credentials
- Human approval checkpoints and audit of every action
Decisions to make
- Single agent with tools, planner-executor, or supervised multi-agent
- Which actions are autonomous and which need approval
- Whose identity an agent acts under
Warning sign: Agents with broad service-account credentials that can do more than the user who asked.
Floor 3
Models and gateway
One controlled path between every application and every model, commercial or open-weight.
How we design this floorCore components
- LLM gateway with routing, fallbacks and quotas
- Guardrails and policy enforcement on inputs and outputs
- Evaluation harness and release gates for model and prompt changes
Decisions to make
- Which models are approved for which data classes
- Build or buy the gateway
- How cost is attributed to teams and use cases
Warning sign: Every team holding its own API keys, prompts and spend, invisible to security and finance.
Floor 2
Knowledge and retrieval
Grounds answers in the organization’s own content and returns only what the user is allowed to see.
How we design this floorCore components
- Ingestion and parsing pipelines with incremental sync
- Hybrid (vector plus keyword) search index
- Permission-aware retrieval and citation of sources
Decisions to make
- Chunking and parsing strategy by document type
- How source permissions are enforced at query time
- How answer quality is measured and monitored
Warning sign: A vector index built from a one-time export that ignores who is allowed to read what.
Floor 1
Data foundation
Trustworthy, governed data products that every floor above depends on.
How we design this floorCore components
- Lakehouse or warehouse with governed data products
- Data contracts with the systems that produce data
- Lineage, quality monitoring and classification
Decisions to make
- Which data products priority use cases need first
- Who owns each source and its quality
- How sensitive data is classified and kept out of reach
Warning sign: Pilots built on a clean extract that production data never matches.
The core: governance and security through every floor
- Identity
- Every request and every agent action carries a real identity, and nothing exceeds the permissions of the person it acts for.
- Policy
- Approved models, data classes and use cases are enforced in the platform, not only written in a document.
- Audit trail
- Prompts, retrievals, tool calls and outputs are logged so any answer or action can be reconstructed.
- Observability
- Quality, latency, cost and drift are monitored in production, with thresholds that trigger review.
- Cost control
- Spend is attributed to teams and use cases, with quotas and routing that keep it predictable.
Design the core against recognized frameworks: the NIST AI RMF, ISO/IEC 42001 and the OWASP Top 10 for LLM Applications.
Use and cite the framework
Score your organization against it with the free AI readiness assessment, or read the component detail in the enterprise AI reference architecture guide.
Enterprise AI Architects. "The Core-and-Floors Model: an enterprise AI reference architecture framework." 2026. https://enterpriseaiarchitects.com/framework
About the Core-and-Floors Model
What is the Core-and-Floors Model?
The Core-and-Floors Model is the enterprise AI reference architecture framework Enterprise AI Architects uses with clients. It describes production AI as five floors (data foundation, knowledge and retrieval, models and gateway, agents and orchestration, experience) resting on each other, with a governance and security core (identity, policy, audit, observability, cost control) running through every floor.
Why is it called Core-and-Floors?
The name comes from how tall buildings are engineered: a structural core carries load through every floor. In the model, governance and security are the core, and the five floors are data foundation, knowledge and retrieval, models and gateway, agents and orchestration, and experience. Enterprise AI Architects first published it in October 2026.
Why model enterprise AI as floors and a core?
Because the dependencies run one way. An agent is only as safe as the gateway and identity beneath it, and retrieval is only as good as the data under it. Treating governance and security as a core rather than a top layer makes clear they have to be designed into every floor, the way a building’s core carries load through every storey.
How do I use the framework to find where to start?
Score each floor and the core for the use cases you care about, then start with the weakest one those use cases depend on. Our free AI readiness assessment gives a quick first read, and the enterprise AI reference architecture guide describes each component in more depth.
Can I reuse or cite the Core-and-Floors Model?
Yes. Cite it as: Enterprise AI Architects, "The Core-and-Floors Model: an enterprise AI reference architecture framework," with a link to this page. You may reproduce the diagram with attribution.
Bring us the architecture question you are stuck on.
Kelsey, our AI assistant, can answer questions and book you a 30-minute briefing with a senior architect in about two minutes, by voice or by typing.
- Vendor-neutral advice. We resell nothing.
- Senior architects on the call, not a sales team.
- Mutual NDA available before we talk specifics.
Book with Kelsey
She finds a time that works, books it on our calendar and sends the invite. No forms.
By talking or chatting with Kelsey, you agree we may text and email you about your request. Message and data rates may apply. Reply STOP to opt out. Terms and Privacy.
Prefer a person? Call (301) 301-5765.