AGENTIC WORKFLOWS

█

Repeatable interaction patterns for AI agents in industrial automation.

Agent workflows define how agents think, act, and decide when analyzing PLC code, diagnosing faults, or validating safety logic. They're not just prompts—they're structured execution patterns with explicit boundaries, tool policies, and human approval gates.

From single-agent advisory analysis to distributed multi-facility coordination, these canonical patterns guide you from safe experimentation to production deployment.

PATTERNS

8 workflows

Canonical patterns

TRACKS

4 levels

Technician → Architect

MATURITY

Production

Battle-tested

WHAT ARE AGENTIC WORKFLOWS?

Workflows define the complete execution pattern for an agent—not just what it does, but when it stops, what it can't access, and where humans must approve.

Each workflow specifies:

→ Trigger rules (manual, scheduled, event-driven)
→ Reasoning steps (ReAct, Chain-of-Thought, debate)
→ Safety boundaries (read-only vs write-enabled)
→ Human gates (mandatory approval points)
→ Stop conditions (confidence thresholds, max iterations)

Generic Workflow Shape

      graph LR
    A[Trigger] --> B[Observe]
    B --> C[Reason]
    C --> D[Structure]
    D --> E[Human Review]
    E --> F[Stop]

    style A fill:#04d9ff,stroke:#04d9ff,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#04d9ff
    style C fill:#1a1a1e,stroke:#04d9ff
    style D fill:#1a1a1e,stroke:#04d9ff
    style E fill:#fec20b,stroke:#fec20b,color:#0d0d0f
    style F fill:#1a1a1e,stroke:#04d9ff

⚠️

⚠️ SAFETY BOUNDARY REMINDER

This tutorial performs analysis only.

It must never be connected to:

Live PLCs
Production deployment pipelines
Safety-rated controllers
Motion or power systems

> All outputs are advisory-only and always require explicit human approval before any real-world action.

PATTERN LIBRARY

Filter by:

Single Agent

WF-Advisory-Analysis

Single-Run Advisory Reasoning with Human Review

🟢 Technician Risk: Low

      graph LR
    A[Trigger] --> B[Observe]
    B --> C[Reason]
    C --> D[Structure]
    D --> E[Human Review]
    E --> F[Stop]

    style A fill:#10809c,stroke:#10809c,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#10809c
    style C fill:#1a1a1e,stroke:#10809c
    style D fill:#1a1a1e,stroke:#10809c
    style E fill:#fec20b,stroke:#fec20b,color:#0d0d0f
    style F fill:#1a1a1e,stroke:#10809c

Use Cases

→ PLC/ST logic review (missing guards, unclear states)
→ Alarm log diagnosis (correlate sequences to fault patterns)
→ Code extraction (I/O lists, alarms, parameters to JSON)

Coming Soon

Single Agent

WF-ReAct-ReadOnly

Reasoning + Acting with Read-Only Tools

🟢 Technician Risk: Low

      graph LR
    A[Start] --> B[Thought]
    B --> C[Action]
    C --> D[Observation]
    D --> E{Sufficient?}
    E -->|No| B
    E -->|Yes| F[Final Answer]
    F --> G[Human Review]

    style A fill:#10809c,stroke:#10809c,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#10809c
    style C fill:#1a1a1e,stroke:#9e4aff
    style D fill:#1a1a1e,stroke:#10809c
    style E fill:#fec20b,stroke:#fec20b,color:#0d0d0f
    style F fill:#1a1a1e,stroke:#10809c
    style G fill:#fec20b,stroke:#fec20b,color:#0d0d0f

Use Cases

→ Multi-file code investigation
→ Diagnostic evidence gathering across logs
→ Cross-referencing documentation and code

Coming Soon

Multi Agent

WF-Coordinator-Worker

Multi-Agent Coordination with Task Distribution

🟣 Developer Risk: Medium

      graph TD
    A[Coordinator] --> B[Worker 1]
    A --> C[Worker 2]
    A --> D[Worker 3]
    B --> E[Synthesis]
    C --> E
    D --> E
    E --> F[Human Review]

    style A fill:#9e4aff,stroke:#9e4aff,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#04d9ff
    style C fill:#1a1a1e,stroke:#04d9ff
    style D fill:#1a1a1e,stroke:#04d9ff
    style E fill:#1a1a1e,stroke:#9e4aff
    style F fill:#fec20b,stroke:#fec20b,color:#0d0d0f

Use Cases

→ Large codebase analysis (parallel file reviews)
→ Multi-system validation (PLC + HMI + safety logic)
→ Consensus-based code review

Coming Soon

Multi Agent

WF-Validation-Swarm

Multi-Perspective Validation with Voting

🟣 Developer Risk: Medium

      graph TD
    A[Input] --> B[Safety Agent]
    A --> C[Performance Agent]
    A --> D[Standards Agent]
    B --> E[Voting/Consensus]
    C --> E
    D --> E
    E --> F[Aggregated Result]
    F --> G[Human Review]

    style A fill:#04d9ff,stroke:#04d9ff,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#00ff7f
    style C fill:#1a1a1e,stroke:#9e4aff
    style D fill:#1a1a1e,stroke:#fec20b
    style E fill:#1a1a1e,stroke:#9e4aff
    style F fill:#1a1a1e,stroke:#9e4aff
    style G fill:#fec20b,stroke:#fec20b,color:#0d0d0f

Use Cases

→ Safety-critical code validation (IEC 61508)
→ Multi-criteria design review
→ Compliance checking (multiple standards)

Coming Soon

Production

WF-Event-Driven-Monitor

Automated Trigger with Shadow Mode Deployment

🔵 System Risk: Medium

      graph LR
    A[Event Source] --> B[Trigger]
    B --> C[Workflow Execution]
    C --> D[Shadow Mode Decision]
    D --> E[Log Result]
    E --> F[Dashboard/Alerts]
    F --> G[Human Reviews Logs]

    style A fill:#1f75ff,stroke:#1f75ff,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#1f75ff
    style C fill:#1a1a1e,stroke:#1f75ff
    style D fill:#1a1a1e,stroke:#fec20b
    style E fill:#1a1a1e,stroke:#1f75ff
    style F fill:#1a1a1e,stroke:#1f75ff
    style G fill:#fec20b,stroke:#fec20b,color:#0d0d0f

Use Cases

→ Automated alarm diagnosis on trigger
→ Scheduled code quality scans
→ Continuous compliance monitoring

Coming Soon

Production

WF-Shadow-Mode-Monitor

Production Deployment Without Actuation

🔵 System Risk: Low

      graph LR
    A[Production Data] --> B[Agent Workflow]
    B --> C[Generate Recommendation]
    C --> D[Log Decision]
    D --> E[Monitoring Dashboard]
    E --> F[Human Reviews]
    F --> G{Approve?}
    G -->|Yes| H[Manual Action]
    G -->|No| I[Ignore]

    style A fill:#1f75ff,stroke:#1f75ff,color:#0d0d0f
    style B fill:#1a1a1e,stroke:#1f75ff
    style C fill:#1a1a1e,stroke:#1f75ff
    style D fill:#1a1a1e,stroke:#1f75ff
    style E fill:#1a1a1e,stroke:#1f75ff
    style F fill:#fec20b,stroke:#fec20b,color:#0d0d0f
    style G fill:#fec20b,stroke:#fec20b,color:#0d0d0f
    style H fill:#00ff7f,stroke:#00ff7f,color:#0d0d0f
    style I fill:#1a1a1e,stroke:#64748b

Use Cases

→ Pre-production agent validation
→ Continuous compliance monitoring
→ Safety-critical advisory systems

Coming Soon

Distributed

WF-Federated-MCP

Cross-Facility MCP Server Coordination

🔴 Architect Risk: High

      graph TB
    A[Facility 1 MCP] --> D[Federation Layer]
    B[Facility 2 MCP] --> D
    C[Facility 3 MCP] --> D
    D --> E[Coordinated Workflow]
    E --> F[Distributed Results]
    F --> G[Local Human Review]

    style A fill:#ff4fd8,stroke:#ff4fd8,color:#0d0d0f
    style B fill:#ff4fd8,stroke:#ff4fd8,color:#0d0d0f
    style C fill:#ff4fd8,stroke:#ff4fd8,color:#0d0d0f
    style D fill:#1a1a1e,stroke:#ff4fd8
    style E fill:#1a1a1e,stroke:#ff4fd8
    style F fill:#1a1a1e,stroke:#ff4fd8
    style G fill:#fec20b,stroke:#fec20b,color:#0d0d0f

Use Cases

→ Multi-plant root cause analysis
→ Cross-facility best practice mining
→ Federated pattern learning

Coming Soon

Coordination

WF-Hierarchical-Coordination

Multi-Level Agent Orchestration

🔴 Architect Risk: High

      graph TD
    A[Meta-Coordinator] --> B[Coordinator 1]
    A --> C[Coordinator 2]
    B --> D[Worker 1a]
    B --> E[Worker 1b]
    C --> F[Worker 2a]
    C --> G[Worker 2b]
    D --> H[Synthesis L2]
    E --> H
    F --> I[Synthesis L2]
    G --> I
    H --> J[Meta-Synthesis]
    I --> J
    J --> K[Human Review]

    style A fill:#ff4fd8,stroke:#ff4fd8,color:#0d0d0f
    style B fill:#9e4aff,stroke:#9e4aff,color:#0d0d0f
    style C fill:#9e4aff,stroke:#9e4aff,color:#0d0d0f
    style D fill:#1a1a1e,stroke:#04d9ff
    style E fill:#1a1a1e,stroke:#04d9ff
    style F fill:#1a1a1e,stroke:#04d9ff
    style G fill:#1a1a1e,stroke:#04d9ff
    style H fill:#1a1a1e,stroke:#9e4aff
    style I fill:#1a1a1e,stroke:#9e4aff
    style J fill:#1a1a1e,stroke:#ff4fd8
    style K fill:#fec20b,stroke:#fec20b,color:#0d0d0f

Use Cases

→ Full-plant digital twin validation
→ Enterprise-wide code modernization
→ Multi-system integration projects

Coming Soon

COMPARE WORKFLOWS

Choose the right pattern for your use case

Pattern	Track	Complexity	Latency	Cost/Run	Tutorials
WF-Advisory-Analysis	🟢 Technician	Low	2-8 seconds	$0.01-0.05	4 tutorials
WF-ReAct-ReadOnly	🟢 Technician	Medium	5-15 seconds	$0.03-0.10	2 tutorials
WF-Coordinator-Worker	🟣 Developer	High	10-30 seconds	$0.15-0.50	2 tutorials
WF-Validation-Swarm	🟣 Developer	High	15-45 seconds	$0.20-0.80	1 tutorials
WF-Event-Driven-Monitor	🔵 System	High	2-10 seconds (per event)	$0.02-0.15	2 tutorials
WF-Shadow-Mode-Monitor	🔵 System	Medium	2-10 seconds	$0.02-0.10	2 tutorials
WF-Federated-MCP	🔴 Architect	High	30-120 seconds	$0.50-2.00	2 tutorials
WF-Hierarchical-Coordination	🔴 Architect	High	60-300 seconds	$1.00-5.00	1 tutorials

HOW PATTERNS EVOLVE

Example: WF-Advisory-Analysis across all 4 tracks

🟢 Technician

└─> Single-agent, single-run, manual trigger

└─> Human review required for all outputs

🟣 Developer

└─> Multi-agent coordination (Coordinator–Worker)

└─> Parallel analysis with consensus reporting

└─> Still advisory-only; human review remains

🔵 System & Deployment

└─> Automated triggers (scheduled, event-driven)

└─> Shadow mode deployment (non-acting monitor)

└─> Production observability and tracing

└─> Still no autonomous action

🔴 Architect

└─> Distributed across teams/facilities

└─> Federated MCP servers

└─> Cross-facility tool discovery

└─> Human oversight remains at decision points

→ Start with Technician Track

READY TO BUILD?

📚

Learn the Basics

Start with Track 1 tutorials to master single-agent workflows and advisory-only patterns.

→ TECHNICIAN TRACK

🔬

Explore Patterns

Browse the complete workflow library and find the right pattern for your use case.

→ PATTERN LIBRARY