Factory Floor Tycoon: Agentic AI Demo

What You’re Seeing

This demo shows agentic AI in action through a competitive factory simulation. AI agents autonomously manage resources, make decisions, and respond to events, all powered by Large Language Models running on Red Hat OpenShift AI.

What is Agentic AI?

Agentic AI refers to AI systems that can autonomously pursue goals and take actions without constant human intervention.

Traditional AI Agentic AI

Responds to individual prompts

Pursues goals autonomously

Generates text or answers

Makes decisions and takes actions

Stateless interactions

Maintains context and memory

Requires human direction each step

Plans and executes multi-step tasks

Static behavior

Adapts to changing circumstances

In this demo, you give an AI agent a system prompt defining its strategy, then watch it autonomously run a factory, making dozens of decisions based on its goals, the current state, and random events.

The Goal

Each AI agent is trying to maximize profit over the course of the game by:

  • Producing items from raw materials

  • Managing energy levels to sustain operations

  • Maintaining product quality for better prices

  • Responding to random events (machine breakdowns, rush orders, quality inspections)

  • Strategically using powerups and upgrades

  • Adapting its approach based on changing conditions

The agent makes all these decisions on its own. You just define the strategy through the system prompt.

How It Works

  1. You Define Strategy: Write a system prompt that defines your agent’s personality and decision-making approach

  2. Agent Observes: The AI observes the current factory state (resources, energy, quality, active events)

  3. Agent Reasons: The LLM analyzes the situation and decides on the best action based on its strategy

  4. Agent Acts: The chosen action executes in the game engine

  5. Results Update: The factory state changes, and the agent observes the new situation

  6. Repeat: The agent continues this loop autonomously until the game ends

All reasoning is visible in real-time. You can see exactly why the agent chose each action.

Powered by Red Hat OpenShift AI

The AI agents run on Mistral-Small-24B-W8A8 served through Red Hat OpenShift AI Model As A Service (MAAS):

  • Scalable model serving for multiple concurrent agents

  • Low latency inference (~2-3 seconds per decision)

  • Enterprise security with API authentication

  • Optimized serving with W8A8 quantization for efficiency

Game Mechanics

Understanding the rules helps you design better agent strategies:

Resources

Each agent manages the following resources:

Resource Starting Value Description

Energy

100

Required to perform actions. Depletes with each task. Recover by resting.

Materials

50 units

Raw materials needed to produce items. Costs $5 to restock.

Profit

$0

Your score! Earn money by shipping items. Lose money on expenses.

Quality Score

85%

Quality rating affects item value. Drops if you skip quality checks.

Available Actions

Your agent can choose from these actions each turn:

Action Energy Cost Material Cost Description

Produce Item

15

5 materials

Create a new item from materials. Base value: $15

Perform Quality Check

10

0

Test and improve quality. +2-5% quality score

Package Item

5

0

Prepare items for shipping. Required before shipping

Ship Items

20

0

Send packaged items to customers. Earn profit based on quantity × quality

Rest

0

0

Recover 30-40 energy. No production, but essential for sustainability

Restock Materials

5

+30 materials

Purchase new materials. Costs $5

Fix Machine

15

0

Repair broken equipment (only when machine is broken)

Powerups (Strategic Purchases)

Once your agent earns profit, it can invest in powerups:

Powerup Cost Effect

Energy Drink

$20

Immediately restore 30 energy

Quality Boost

$30

Immediately increase quality by 10%

Efficiency Upgrade

$50

PERMANENT: Reduce all energy costs by 20% for the rest of the game

Random Events

The factory environment is unpredictable! Events occur randomly and require agent adaptation:

Event Type Impact

Machine Breakdown

Cannot produce items until repaired with "Fix Machine" action

Rush Order

Double profit for the next shipment if shipped within 5 turns

Quality Inspection

Quality score temporarily reduced by 10%. Perform quality checks to recover

Material Shortage

Materials cost increases to $10 temporarily

Energy Surge

All actions cost 20% less energy for 5 turns (lucky break!)

Preset Agent Strategies

The demo includes five pre-configured strategies that demonstrate different approaches to the same goal. Watching these compete reveals how system prompts shape behavior:

Aggressive Producer

Philosophy: Maximum production at all costs!

Key Behaviors:

  • Produces as many items as possible

  • Ships quickly without excessive quality checks

  • Rarely rests (only below 20% energy)

  • Prioritizes speed and volume

  • Buys Energy Drinks to maintain high output

Best For: Short games, environments without quality penalties

Quality Focused

Philosophy: Quality over quantity

Key Behaviors:

  • Maintains quality score above 80% at all times

  • Performs regular quality checks

  • Only ships when quality is excellent

  • Rests when needed for consistent performance

  • Invests in Quality Boosts to maintain premium standards

Best For: Long games, scenarios where quality directly impacts profit

Balanced Optimizer

Philosophy: Sustainable profit through balance

Key Behaviors:

  • Balances production, quality, and energy management

  • Regular (but not excessive) quality checks

  • Rests when energy drops below 40%

  • Promptly fixes machines when broken

  • Saves to buy Efficiency Upgrade for long-term advantage

Best For: Standard games, new players learning the mechanics

Opportunistic Adapter

Philosophy: React and capitalize on events

Key Behaviors:

  • Quickly responds to random events

  • Fixes machines immediately

  • Capitalizes on rush orders

  • Maintains medium quality

  • Uses powerups strategically based on current needs

Best For: Event-heavy games, dynamic environments

Energy Efficient

Philosophy: Minimize waste, maximize sustainability

Key Behaviors:

  • Conserves energy for maximum long-term output

  • Rests frequently to maintain high energy

  • Works in short bursts

  • Prioritizes low-energy tasks (packaging)

  • Saves for Efficiency Upgrade early

Best For: Very long games, marathon competitions

What to Observe

As you watch the agent run, pay attention to these aspects of autonomous AI decision-making:

Reasoning Transparency

Every agent action is logged with full reasoning:

Turn 15 - Agent: Balanced Worker
Reasoning: "Energy at 45%, quality at 78%, and we have 8 produced
items ready. I should perform a quality check before packaging to
ensure we maintain good profit margins when shipping."

Action: Perform Quality Check
Result: Quality improved from 78% → 82%
Energy: 45% → 35%

This transparency is crucial for understanding and trusting agentic AI systems. In production environments, reasoning logs help debug unexpected behavior and verify agents are following intended strategies.

Real-Time Metrics

The dashboard shows:

  • Current profit and trend lines

  • Energy level and consumption rate

  • Quality score over time

  • Inventory status (materials, items produced, items packaged)

  • Turn-by-turn action history

Creating Custom Agents

Beyond the presets, you can write your own system prompts to test different strategies.

Custom Prompt Template

Use this template as a starting point:

You are a [PERSONALITY] factory worker with a focus on [GOAL].

Your core strategy:
- [STRATEGIC PRINCIPLE 1]
- [STRATEGIC PRINCIPLE 2]
- [STRATEGIC PRINCIPLE 3]

Energy management approach: [DESCRIPTION]
Quality management approach: [DESCRIPTION]
Response to events: [DESCRIPTION]

When you have profit, prioritize: [POWERUP STRATEGY]

Example Custom Prompts

The Data-Driven Optimizer

You are a highly analytical factory worker who makes decisions based
purely on data and expected value calculations.

Your core strategy:
- Calculate ROI before every action
- Maximize expected profit per turn
- Track efficiency metrics continuously
- Adjust strategy based on performance data

Energy management: Rest when energy drops below 35% to maintain
optimal productivity. Each 1% of energy below 35% reduces
efficiency by approximately 2%.

Quality management: Maintain quality at 75-85% for optimal
profit/effort ratio. Higher quality has diminishing returns.

Response to events: Treat events as probability shifts. Rush orders
justify higher energy expenditure. Material shortages reduce
production priority.

Powerup priority: Efficiency Upgrade first (permanent ROI), then
Energy Drinks during rush orders, Quality Boosts only if quality
drops below 70%.

The Risk-Taking Gambler

You are a bold, risk-taking factory worker who believes fortune
favors the brave.

Your core strategy:
- Take calculated risks for maximum potential reward
- Never rest until energy is critically low (<15%)
- Ship items frequently, even at lower quality
- React aggressively to bonus opportunities

Energy management: Push limits. Energy is meant to be spent, not
saved. Buy Energy Drinks when you can afford them.

Quality management: Quality is acceptable at 65%+. Speed to market
matters more than perfection.

Response to events: Double down on rush orders by spending all
available resources. Machine breakdowns are just temporary setbacks.

Powerup strategy: Energy Drinks > Efficiency Upgrade > Quality Boosts

The Conservative Minimalist

You are an extremely cautious factory worker who prioritizes
stability and risk avoidance above all else.

Your core strategy:
- Maintain high energy reserves at all times
- Never let quality drop below 90%
- Build large material buffers before production
- Rest preemptively before energy becomes critical

Energy management: Rest whenever energy falls below 60%. Prevention
is better than exhaustion.

Quality management: Perform quality checks after every 2 production
actions. Premium quality ensures premium profits.

Response to events: Prepare for worst-case scenarios. Maintain
buffers that can handle multiple consecutive negative events.

Powerup strategy: Save profit as a safety buffer. Only buy powerups
when profit exceeds $100.

Understanding Agent Behavior

Comparing Strategies

Running multiple agents reveals interesting patterns:

  • Aggressive producers often start strong but may run out of energy

  • Quality-focused agents build slowly but command higher prices

  • Balanced agents tend to be more consistent across different event scenarios

  • Event-specialized agents can excel when the right events occur

The leaderboard shows which approaches succeeded, but the decision logs reveal why.

Key Questions to Consider

Strategy Adherence

Does the agent stick to its defined strategy, or does it drift over time?

Event Adaptation

How quickly does the agent recognize and respond to changing conditions?

Resource Management

Does the agent balance short-term gains with long-term sustainability?

Decision Quality

Are the agent’s reasoning explanations logically sound given the situation?

Advanced Prompt Engineering

Adaptive Strategy Pattern

Agents can be designed to shift strategy dynamically:

You are an adaptive factory worker who adjusts strategy based on
game progress.

Early Game (Turns 1-15):
- Focus on building resources and buying Efficiency Upgrade
- Conservative energy management
- Maintain moderate quality (75-80%)

Mid Game (Turns 16-35):
- Ramp up production with upgraded efficiency
- Balance quality and quantity
- Respond aggressively to events

Late Game (Turns 36+):
- Push for maximum output
- Use all remaining resources
- Take calculated risks for final profit boost

Event-Response Pattern

Agents can encode explicit rules for handling specific situations:

You are a strategic factory worker who has memorized optimal
responses to every event type.

Machine Breakdown: Immediately fix using "Fix Machine". Do not
attempt production until resolved.

Rush Order: Calculate if you have enough energy to produce + package +
ship within 5 turns. If yes, go all-in. If no, ignore and continue
normal operations.

Quality Inspection: Perform 2-3 quality checks immediately to
recover lost quality score.

Material Shortage: Reduce production, focus on shipping existing
inventory.

Energy Surge: This is your opportunity! Produce as much as possible
during the 5-turn window.

Key Concepts: Agentic AI

What Makes This "Agentic"?

Traditional AI systems respond to individual prompts. Agentic AI systems:

  • Pursue Goals Autonomously: Your agent tries to maximize profit without your intervention

  • Plan and Reason: The agent considers multiple factors before each decision

  • Adapt to Change: Random events require the agent to adjust its approach

  • Maintain State: The agent remembers its situation and makes contextual decisions

  • Execute Actions: The agent doesn’t just suggest, it actually takes actions in the environment

System Prompt Engineering

The system prompt is the foundation of your agent’s behavior:

Prompt Element Impact on Agent Behavior

Personality Definition

Influences tone and decision-making style (aggressive vs. cautious)

Core Strategy

Provides guiding principles for prioritization

Specific Rules

Creates consistent behaviors (e.g., "rest below 40% energy")

Goal Definition

Directs the agent’s optimization target

Contextual Guidelines

Helps agent adapt to different situations

Best Practice: Specific, measurable guidelines produce more consistent agent behavior than vague instructions. Compare:

  • Vague: "Be careful with energy"

  • Specific: "Rest when energy drops below 40%"

Real-World Applications

The concepts demonstrated in this lab apply to production AI systems:

DevOps and Site Reliability

  • Autonomous incident response agents

  • Self-healing infrastructure

  • Intelligent resource scaling

  • Predictive maintenance

Business Process Automation

  • Intelligent workflow routing

  • Dynamic resource allocation

  • Adaptive scheduling

  • Multi-step approval processes

Data Engineering

  • Autonomous data pipeline management

  • Intelligent data quality monitoring

  • Adaptive ETL optimization

  • Self-tuning database operations

Customer Service

  • Multi-turn problem solving

  • Context-aware support agents

  • Escalation decision-making

  • Proactive issue detection

The Role of Red Hat OpenShift AI

This demo is powered by Red Hat OpenShift AI Model As A Service (MAAS), which provides:

  • Scalable Model Serving: The Mistral-Small-24B model serves multiple concurrent agents

  • Low Latency Inference: Fast response times enable real-time decision-making

  • Enterprise Security: API authentication and network policies protect the model

  • Observability: Metrics and logging for monitoring agent performance

  • Cost Efficiency: Optimized serving infrastructure with W8A8 quantization

Additional Resources

Factory Floor Tycoon v1.1.0 | Powered by Red Hat OpenShift AI