How Ontology‑Driven Harness Engineering Enables Controllable Agent Execution

01. From the Agent Hype to "Uncontrollable"

In 2024‑2025, agents become the main form of enterprise AI, capable of planning, tool invocation, and multi‑step execution. In demos they shine, but in production they frequently misuse terminology, deviate in reasoning, and produce results that violate corporate policies, essentially "confidently doing the wrong thing".

The root cause is not model weakness but the lack of a "rule‑aware structure"—agents know how to act but not the business boundaries or decision rules that constrain them.

Harness Engineering emerges as a response, aiming to embed constraints directly into the business structure rather than as external engineering patches.

02. Redefining the Problem: "Safe and Controllable" as a Multi‑Dimensional Engineering Question

Before proposing a solution, the article clarifies the dimensions of safe, controllable execution:

Permission & Isolation : Who can do what? Can data cross domains? (RBAC/ABAC, API gateways, data sandboxes)

Behavior Constraints : Where are the agent's reasoning and invocation limits? (Prompt constraints, tool whitelists, ontology modeling)

Audit & Traceability : What was done? Can the decision process be reconstructed? (Operation logs, decision‑chain tracing, explainability frameworks)

Exception Handling : How to degrade or roll back on errors? (Circuit breakers, manual review nodes, idempotent design)

Result Validation : Does the output meet business rules? (Rule engines, formal verification, ontology‑based constraint checks)

Compliance Alignment : Does it satisfy industry regulations? (Compliance knowledge base, approval flow integration, auditable reports)

The ontology‑driven approach focuses on the "Behavior Constraints" and "Result Validation" dimensions, providing a semantic infrastructure layer that replaces ad‑hoc prompts and rule scripts.

03. Architectural Constraints: From "External Fences" to "Built‑In Skeleton"

Traditional engineering constraints work in simple scenarios but face three structural challenges in complex business:

Rule count grows with business complexity, inflating maintenance cost.

Rules expressed in natural language are ambiguous, leading to unreliable model interpretation.

Semantic relationships between rules and business objects are implicit, hindering reuse across scenarios.

Ontology changes this by defining the agent's action space within the business structure itself. Constraints become part of the ontology, not external prompts, and are stored as queryable, verifiable structures.

During execution, the agent's intent is compared against the ontology after generation; violations trigger a deterministic rollback rather than silent acceptance.

04. Context Engineering: From "Memory Augmentation" to "Reconstructing Memory"

Agents often lose context in long tasks, repeatedly asking for basic information or dropping critical details. The article argues that this stems from linear text stacking without structure.

Ontology provides a graph‑based business semantic network, enabling three concrete improvements:

Precise Retrieval Instead of Full Injection : Before an agent runs, the cognition engine extracts a relevant sub‑graph from the ontology and injects only that context, eliminating overflow and irrelevant data.

Consistency Assurance : A unified semantic network allows systematic handling of outdated, conflicting, or redundant information, ensuring the agent always reasons on the latest consistent data.

Cross‑Task Reuse : The same semantic structure can serve multiple agents without rebuilding context for each scenario.

The ontology also bridges the gap between symbolic reasoning (knowledge graphs) and LLM inference: where the ontology covers a domain, it enforces deterministic constraints; where it does not, the LLM fills the gap with confidence‑annotated results.

05. Feedback Loop: From "Subjective Evaluation" to "Traceable Verification"

Current feedback mechanisms rely on a secondary evaluator model, which can be fooled by superficially plausible outputs and lacks business‑level judgment.

The ontology‑driven loop replaces subjective assessment with objective, structure‑based verification. Business judgments—such as quota limits, prerequisite checks, or workflow logic—are encoded as rules that can be automatically validated against each agent output.

Hard constraints are directly enforceable; soft constraints combine LLM assessment or human review. The loop also enables continuous ontology evolution: agents expose uncovered concepts and frequent error paths, guiding ontology refinement.

Each verification step produces an auditable trace, satisfying regulatory needs in highly compliant industries.

06. From Technical Controllability to Business Controllability: The Knora Implementation Path

Knora, a platform from 悦点科技, materializes the methodology into a layered system:

Ontology Layer (Knowledge Base) : Stored as a labeled‑property graph (LPG) with five core concepts—Entity, Relation, Event, Action, Logic (DAG‑based execution flow).

Cognitive Engine (Translation & Arbitration) : Before agent launch, it queries the ontology for relevant entities, relations, rules, and tools, injecting them into the agent's reasoning context. After generation, it validates results against the ontology and forces re‑reasoning on violations.

Agent Execution Layer : Receives user tasks, calls tools, and produces results, but all tool usage, triggers, and flows are dictated by the ontology.

Data flow: User task → Cognitive engine extracts relevant knowledge → Agent reasons & executes → Result returns to cognitive engine for ontology validation → If passed, output is delivered; otherwise, rollback and error handling occur.

Concrete example: a work‑order quantity change from 500 to 800 exceeds a 20% approval threshold. The ontology defines the work‑order entity, the change event, the required approval action, and the relationship chain to approvers. The system blocks the direct write, generates a structured error report, creates an approval task, logs the intent, and only after approval updates the work order.

Knora’s automatic modeling strategy balances full automation for well‑structured tasks (e.g., field‑to‑ontology mapping, DAG generation) with human‑in‑the‑loop verification for ambiguous semantic judgments, using confidence scores to route items to manual review.

07. Real‑World Deployments and Impact

Knora is already deployed in energy transport, electronics manufacturing, finance, and security:

In railway inspection, a smart agent reduced a 30‑person‑week data‑to‑report process to a 3‑person‑day workflow, achieving >70× efficiency gains across dozens of scenarios.

In electronics, quality traceability and defect analysis have been transformed from experience‑driven manual steps to precise, knowledge‑driven automation.

The article concludes that enterprise AI competition will shift from model superiority to who can first structure their business knowledge into an ontology, creating a self‑evolving intelligent foundation that outlives any specific model or tool.