Which parts are deterministic and which are agentic?
Words made simple
A few useful meanings before you read the explanation.
- Deterministic
- Following steps or rules defined by code. The model does not decide which step should run next in a deterministic workflow.
- LLM
- A model trained to work with language. It can write or summarize text, but its answer still needs checking against reliable information.
- Retrieval
- Finding and fetching information that is useful for the current question.
More words from this lesson
- Grounding
- Connecting an answer to supplied evidence, such as a policy document or a verified database result.
- Agent
- An application that uses a model to choose its next action within defined limits. It can use tools and continue through several steps.
- Durable
- Able to retain progress and recover across interruptions. Durable workflows require appropriate storage and retry behavior.
- Idempotency
- Making a repeated request have the same intended business effect as one request. It helps avoid duplicate submissions or payments.
- Authorization
- Checking what that identity is allowed to read or do. Being logged in does not grant access to every record.
| Responsibility | Approach | Why |
|---|---|---|
| Login, tenant/customer identity | Deterministic | Permissions cannot depend on a model |
| Policy ownership and active version resolution | Deterministic service lookup | Authoritative applicability must be verified |
| Exact claim status | Deterministic authorized tool | The backend is the source of truth |
| Premium/deductible calculations | Deterministic business rules | Exact numerical/business behavior |
| Policy clause retrieval | Controlled search service | Eligibility, versions, and budgets are explicit |
| Explain retrieved wording | LLM with source grounding | Useful language transformation |
| Interpret accident description | Structured model extraction | Handles varied natural language |
| Ask for missing fields/documents | Rules plus bounded agent wording | Rules know what is required; model communicates |
| Choose among safe information tools | Bounded agent/graph | Next information need may vary |
| Create/update draft | Deterministic record write from validated proposal | Draft lifecycle and concurrency need control |
| Determine required approval | Deterministic policy | Risk controls cannot be negotiated by prompt |
| Submit claim / modify policy | Authorized deterministic workflow | Exact consent, business invariants, idempotency |
| Claim denial or settlement | Authoritative rules and authorized human process | Consequential adjudication is outside casual agent authority |
| Draft email | LLM allowed | Language assistance |
| Send email | Approved recipients/content + deterministic sender | External effect and disclosure risk |
| Retention, deletion, auditing | Deterministic policies/services | Consistent data governance |
Agentic portion: interpreting wording, gathering missing information, and choosing among bounded read/search actions. Workflow portion: identity, required fields, approvals, transaction order, retries, and durable outcomes. LangGraph can coordinate the conversation; a durable business workflow coordinates a multi-day operational claim process if required. MCP is optional when several hosts reuse the tools.
Can you explain it without the notes?
Which parts are deterministic and which are agentic?
Use your own words. Explain an example and one limitation before checking the explanation.
Check your explanation
Agentic portion: interpreting wording, gathering missing information, and choosing among bounded read/search actions. Workflow portion: identity, required fields, approvals, transaction order, retries, and durable outcomes. LangGraph can coordinate the…
Turn this concept into a clear answer.
Practice this conceptAdapted from 13_AI.MD · “Which parts are deterministic and which are agentic?”. Examples are study fragments; framework behavior can depend on your version.