The Building Blocks
RAG, agents, MCP and foundation models are not the same kind of thing. What AI does for the business (capabilities and solutions) is delivered by a logical architecture (patterns, standards, components), which is implemented by products and platforms. Not every solution needs every block.
Security, governance and responsible AI
Business
- 1
AI capabilities
- Generate & summarise
- Converse
- Answer from knowledge
- Understand text & documents
- See & hear
- Predict & detect
- Recommend & optimise
- Act on systems
- 2
AI solutions
- Assistants and copilots
- Task agents
- Intelligent automation
Architecture (logical)
- 3
Design patterns
- Multi-agent system
- Workflow with AI steps
- Human in the loop
- 4
Standards and components
- Model gateway and guardrails
Implementation (physical)
- 5
Products and platforms
AI and data platforms
- Model platforms
- Vector databases
- Agent frameworks and workflow tools
- Lakehouse, warehouse and ML platforms
Enterprise systems
Infrastructure and hosting
- Public cloud
- Private cloud and on-premises
- Compute
- Storage
- Containers
- Networking
AI operations and enablement
Business value
- Productivity and efficiency
- Customer and employee experience
- Faster, better decisions
- New services
- Resilient, secure operations
Items with › open for detail; hover the side lanes for their controls. Choosing the AI for a real problem: AI Solution Advisor.
Common mistakes
- “RAG or an agent?” Different levels: RAG is a pattern, an agent decides steps, MCP is a protocol. They often appear together.
- Fine-tuning to teach facts. Tariffs and terms change: retrieve them.
- An agent where a workflow would do. If the steps are known, use a workflow with AI steps inside it.
- Treating the assistant as the source of truth. The catalog and billing stay the systems of record.