How to Use MiroFish AI for Market Simulation in 2026 (Step-by-Step Guide)

You’ve seen the headlines: “I simulated my market launch with 44 AI agents.“ You’ve probably even read the MiroFish setup guide and got it running locally.

But now comes the real question: what do you actually do with it?

Running MiroFish is the easy part. Getting useful, actionable market intelligence out of it — that’s the skill. I’ve spent the last few months running simulations for multiple product ideas, and in this guide I’ll share exactly how to use MiroFish AI to validate product decisions, test messaging, and uncover blind spots before you write a single line of code.

Before you begin: If you haven’t installed MiroFish yet, start with the MiroFish Setup Guide first, then come back here.

What Is MiroFish AI?

MiroFish is an open-source AI market simulation engine that spawns multiple AI agents representing different market personas — customers, competitors, investors, journalists — and runs a simulated market scenario. Each agent acts based on its persona’s motivations, biases, and knowledge, giving you a surprisingly realistic preview of how a real launch might play out.

Think of it as a focus group with 44 participants, running 24/7, costing pennies per session.

Step 1: Prepare Your Seed Documents

Your simulation is only as good as the context you give it. Before running anything, gather three things:

1. Company & Product Brief

Write a 1–2 page document covering:

2. Market Context

• Competitor landscape — Who are the top 3–5 competitors and their positioning
• Market trends — What’s happening in the space right now
• Regulatory factors — Any compliance or legal considerations

3. Buyer Persona Definitions

Define 3–5 personas that would interact with your product. Be specific:

Persona: "Indie Founder Sarah"
- Runs a 3-person B2B SaaS
- Budget-conscious, technical background
- Has been burned by expensive tools that her team never adopted
- Active on X/Twitter, reads Hacker News and Indie Hackers
- Primary decision driver: ROI and time-to-value
- Secondary concern: ease of onboarding for non-technical team members

Pro tip: The more specific your personas, the more nuanced your simulation results. Generic personas produce generic answers.

Step 2: Write Effective Simulation Prompts

This is where most people get it wrong. A weak prompt produces vague, useless output. A good prompt produces insights you can actually act on.

The Anatomy of a Good Prompt

Here’s a template you can use for any simulation:

[SCENARIO]
You are participating in a market simulation for {Product Name},
a {one-line description}.

[CONTEXT]
Background: {2-3 sentences about the market}
Product details: {key features and positioning}
Target users: {who this is for}

[YOUR ROLE]
You are {Persona Name}, a {short description}.
Your background: {relevant experience / bias}
Your goals: {what you want from a product like this}
Your concerns: {what would stop you from buying}

[TASK]
Review the product description below and respond as {Persona Name}:
1. Would you try this product? Why or why not?
2. What is your biggest concern?
3. What would need to be true for you to become a paying customer?
4. How would you compare this to {competitor}?
5. What would you tell a colleague about this product?

{Product description}

Real Example: SaaS Analytics Tool

Here’s a prompt I used to simulate a launch for a (currently hypothetical) AI analytics product called “DataLens”:

[SCENARIO]
You are participating in a market simulation for "DataLens", an AI-powered
analytics platform that turns raw product data into executive-ready reports in minutes.

[CONTEXT]
Background: Business intelligence is a $30B market dominated by complex
tools like Tableau and Power BI that require dedicated analysts. A new wave
of AI-native tools is making analytics accessible to non-technical teams.
Product details: Natural language querying, automated report generation,
Slack-native alerts, pay-per-report pricing ($5/report, no minimum).
Target users: Product managers, startup founders, and marketing leads at
companies with 10–200 employees.

[YOUR ROLE]
You are "Vikram", Head of Product at a 50-person B2B SaaS company.
Your background: 8 years in product management, burned by expensive tools
your team never used, skeptical of "AI-powered" claims after trying several
duds.
Your goals: Find a way to get weekly product metrics without hiring a data
analyst. Need something my whole team can use without training.
Your concerns: Data accuracy, integration complexity, vendor lock-in,
hidden costs.

[TASK]
Review the description below and respond as Vikram:
1. Would you try DataLens? Why or why not?
2. What is your biggest concern?
3. What would need to be true for you to become a paying customer?
4. How would you compare this to Metabase or Tableau?
5. What would you tell a colleague about DataLens?

Why this works: Vikram has a specific background, specific goals, and specific concerns. The prompt forces the agent to role-play authentically rather than giving a generic “looks great!” response.

Step 3: Run Your First Simulation

Once your prompts are ready, configure your agents and run.

Configuration File

Create a simulation-config.yaml:

simulations:
  - name: "SaaS Launch Validation"
    agents:
      - persona: "Indie Founder"
        count: 10
        model: "gpt-4o-mini"
      - persona: "Enterprise Buyer"
        count: 5
        model: "gpt-4o"
      - persona: "Investor"
        count: 3
        model: "gemini-2.0-flash"
      - persona: "Competitor"
        count: 3
        model: "gpt-4o-mini"
      - persona: "Industry Analyst"
        count: 2
        model: "gpt-4o"
    prompt_file: "prompts/saas-launch.yaml"
    rounds: 3
    output: "results/saas-launch/"

Launch the Simulation

mirofish run --config simulation-config.yaml

What Happens Next

Each agent starts responding in its persona’s voice. You’ll see responses streaming in real-time:

• “Indie Founder” agents will ask about pricing, time-to-value, and integration effort
• “Enterprise Buyer” agents will question security, compliance, and support
• “Investor” agents will probe market size and defensibility
• “Competitor” agents will challenge your differentiation claims

Some responses will surprise you. That’s the point.

Estimated run time: 15–20 minutes for a 44-agent simulation
Estimated cost: ~$0.01 for GPT-4o-mini agents, ~$0.20 for GPT-4o agents

Step 4: Interpret the Results

Not all agent feedback is equally valuable. Here’s how to separate signal from noise.

🟢 What To Pay Attention To

🔴 What To Ignore

• Generic praise (“This looks great!”) — agents are polite by default
• Technical nitpicks from non-technical personas — a CEO persona shouldn’t care about your database choice
• Single outlying opinions — one angry agent out of 44 isn’t a trend
• Overly specific feature requests — agents sometimes invent features that don’t exist in your space

Real Insight From My Simulation

When I ran my own market simulation using MiroFish, the most valuable insight wasn’t about pricing or features — it was about team adoption friction. Agents consistently said:

“This looks useful, but I’d worry about getting my team to actually use it.”

That single pattern directly shaped how I designed visibility.so’s onboarding flow. I realized I needed to invest more in templates and guided setup, not just features.

Read the full story: I Simulated My Market Launch Before Writing a Single Line of Code

3 Real Use Cases for MiroFish

Use Case 1: Validate a SaaS Idea Before Building

Scenario: You have a concept for a niche project management tool for remote design teams.

Setup: 30 agents across 4 personas — designers, design managers, freelancers, agency owners. Present them with a landing page mockup and pricing tiers.

What you’ll learn:

• Does the problem actually resonate with real buyers?
• Is your pricing in the right ballpark?
• Which feature do they care about most?
• What’s the #1 reason they’d choose (or skip) your tool?
• Which segment is most likely to pay first?

Use Case 2: Test a Pricing Change

Scenario: You’re considering moving from $29/mo to $49/mo and want to understand the impact.

Setup: 20 agents matching your current customer profile. Present the new pricing and ask how it changes their decision.

What you’ll learn:

• The real price sensitivity curve for each segment
• Which customer profile churns vs stays
• What perceived value justifies the increase
• How competitor pricing factors into the decision
• Whether a grandfathering strategy would help

Use Case 3: Analyze a Competitor Move

Scenario: A competitor just launched a feature that overlaps with your roadmap. Should you pivot, accelerate, or ignore it?

Setup: 15 agents as your customers, 10 as their customers. Present both products side by side with the new feature.

What you’ll learn:

• Whether customers actually care about the new feature
• Which of your existing features do they value more
• Your defensible moats that competitors can’t easily copy
• How your messaging should shift in response
• Whether it’s a genuine threat or noise

Common Mistakes & How to Fix Them

When to Use Cloud vs Local Models

My rule of thumb: Start with local models to iterate on your prompt structure until the responses feel right. Once your prompt is solid, switch to cloud models for the actual simulation run. The local → cloud pipeline costs you nothing for iteration and gives you confidence in the final output.

Frequently Asked Questions

Ready to Run?

The best time to simulate your market is before you build. The second best time is right now.

1. Install MiroFish (if you haven’t already)
2. Draft your seed documents — company brief, market context, buyer personas
3. Write your simulation prompt using the template from Step 2
4. Start with local models to iterate on your prompt
5. Run the full simulation with cloud models
6. Document every objection and pattern

Your market is already having conversations about your space. With MiroFish, you get to listen before you speak.