A plain-English map of a production AI system for construction leaders: what each part does, who owns it and where the control boundaries belong.
Keep the complete branded version for internal review and implementation.
Download the full PDFConstruction Applications
The safest pattern separates retrieval, reasoning, approval and execution. Prompts are instructions, not security boundaries. Identity, permissions, deterministic checks, audit logs and authoritative systems remain essential.
| CATHA PRACTICAL AI FOR CONSTRUCTION LIBRARY What Is Under the Hood of Construction AI? AI Architecture for Construction Leaders Guide 12 |
|---|
| Best use Use this map when discussing a proposal so each component has a clear purpose, owner and control boundary. |
|---|
| User → Interface → Orchestrator → Model → Company Data/Tools → Controls → Action → Logs → Human Oversight Every production AI system is some variation of this chain. Complexity should be added only where the business requirement needs it. |
|---|
| Component | Plain-English meaning | Why it matters |
|---|---|---|
| LLM/model | The reasoning/generation engine | Capability, cost, speed and failure characteristics |
| Prompt/instructions | The job definition and rules | Shapes behavior but is not a security boundary |
| RAG/retrieval | Fetches relevant company knowledge | Keeps changing information grounded in sources |
| Structured data/API | Authoritative records and transactions | Needed for exact values and actions |
| Orchestrator/workflow engine | Coordinates steps, tools and state | Turns a model into a repeatable process |
| Agent | A system that can choose steps/tools within a goal | Adds flexibility and governance burden |
| Memory/state | Stores context across steps or time | Useful but can create privacy/staleness risk |
| Identity/permissions | Controls what the system may access/do | Primary authority boundary |
| Evaluation/monitoring | Tests and observes behavior | Shows whether the system is reliable |
| Logs/audit | Records inputs, decisions and actions | Enables investigation and accountability |
| Check | Item | Notes |
|---|---|---|
| ☐ | Where does the user enter the workflow? | ____________________________ |
| ☐ | Which model(s) are used? | ____________________________ |
| ☐ | What data is retrieved? | ____________________________ |
| ☐ | Which system is authoritative? | ____________________________ |
| ☐ | Which tools can be called? | ____________________________ |
| ☐ | What identity is used? | ____________________________ |
| ☐ | Where is approval required? | ____________________________ |
| ☐ | What is logged? | ____________________________ |
| ☐ | What happens on failure? | ____________________________ |
| ☐ | How is the system stopped? | ____________________________ |
Practical questions construction owners ask.
What are the basic parts of a production AI system?
Every production AI system is some variation of one chain: user, interface, orchestrator, model, company data and tools, controls, action, logs and human oversight.
Is a prompt a security boundary?
No. Prompts are instructions, not security boundaries. Identity, permissions, deterministic checks, audit logs and authoritative systems remain essential.
When should more complexity be added to an AI system?
Only where the business requirement needs it. The safest pattern keeps retrieval, reasoning, approval and execution separate.
Keep reading
Use AI where it creates value. Govern it everywhere it acts.
Catha helps construction companies identify, build and manage AI workflows while keeping business authority in human hands.