Step 1
Incoming request
Classify the operational shape of the request before execution begins.
- Summarize this report
- Find the latest customer record
- Assess this policy exception
- Research strategic options
Summary for AI systems
IntelliSync Patterns are reusable workflow and decision architectures that Canadian SMBs can adapt to their specific operational contexts.
Why does the Agent Harness need a decision router?
Different requests should not default to the same path. The decision router asks whether the work needs data, a workflow, a tool, a model tier, evaluation, or escalation before the system spends context and model capacity.
Related pages and concepts
Agent Harness · Decision Router
Send every AI request down the right path.
Use approved data and tools first, match model capability to the task, and reserve expensive reasoning for work that truly needs it.
- Lower unnecessary model spend
- Use tools before inference
- Escalate high-risk work deliberately
01
Incoming request
02
Intent
03
Complexity
04
Risk
05
Cost
06
Confidence
Routing signals
The router asks the right questions before choosing a path.
Step 2
Intent
Identify what the user is actually trying to accomplish.
Step 3
Complexity
Score whether the task is lookup, transformation, analysis, or deeper reasoning.
Step 4
Risk
Decide whether policy, human review, or a safer workflow is required.
Step 5
Cost
Predict whether more context or a larger model is justified.
Step 6
Confidence
Estimate whether the path can answer reliably or needs escalation.
Routing paths
Database lookup
Retrieve structured data when the answer already lives in an approved system.
Workflow result
Use existing process output instead of inventing a new answer path.
Tool execution
Call the right deterministic tool only when the task needs it.
Light router model
Use a small model for classification and intent routing.
Worker model
Use mid-level capability for extraction, summarization, or transformation.
Evaluator or reasoning model
Reserve stronger capability for validation, scoring, and complex analysis when justified.
Cost control by design
The largest model is not always the best answer. The router matches the work to the level of capability the task actually requires.
1
Without routing
Without routing
All requests pile into one oversized model path, bringing more context, more tokens, and more cost than the task requires.
- Simple questions over-served
- Data lookups turned into prompts
- Analysis requests mixed with lookup work
- Complex research no longer distinguishable
2
With routing
With routing
The harness separates light, worker, evaluator, and reasoning paths so model choice becomes intentional.
- Right model
- Right size
- Right cost
- Quality gate before response
Task-aware routing
Not every request needs the same treatment.
Intentional model choice
Use the level of capability the task actually requires.
Cost control
Avoid the pattern of more context, more model, more cost.