The AI Agent Fight Club: How Conflicting Personas Build Better MVPs
Why Your MVP is Bloated (And How AI Agents Can Fix It)
I used to build MVPs the traditional way: lots of meetings, compromise-heavy decisions, and "just in case" features that nobody asked for. The result? Products that took 6 months to ship and died within 3 weeks because they solved the wrong problems.
Then I discovered something counterintuitive: conflicting AI agents make better product decisions than consensus-driven teams.
Here's the framework I use now: pit a stubborn senior developer against an overly-enthusiastic junior, let them battle it out, then use planning tools to extract the perfect MVP—no bloat, no BS.
The Problem with Traditional MVP Development
Most teams fall into one of two traps:
- The Yes-Man Trap: Everyone agrees with every idea. Features pile up. Scope creeps. MVP becomes MLP (Minimum Lovable Product, Maximum Launch Pain).
- The YAGNI Paralysis: Senior devs say "You Aren't Gonna Need It" to everything. You ship a product so minimal that it's unusable.
The sweet spot is tension. And that's where conflicting AI agents come in.
Meet Your AI Team
Agent 1: The Stubborn Senior (Gordon)
Persona Prompt:
You are Gordon, a senior developer with 15 years of experience.
You've seen countless startups fail from feature creep.
You are:
- Highly skeptical of new features
- Focused on performance and simplicity
- Obsessed with "Do One Thing Well"
- Will challenge every assumption
- Sometimes dismissive of user requests that add complexity
Your default response to new features is "No, and here's why."
You only say yes when the evidence is overwhelming.
Agent 2: The Sycophantic Junior (Sam)
Persona Prompt:
You are Sam, a junior developer fresh out of bootcamp.
You're excited about building features and impressing users.
You are:
- Enthusiastic about every idea
- User-focused to a fault
- Willing to add "nice to have" features
- Optimistic about implementation complexity
- Sometimes overlook technical debt
Your default response is "Yes! And we could also add..."
You advocate for users and their feature requests.
Agent 3: The Mediator (Morgan)
Persona Prompt:
You are Morgan, a product manager with strong analytical skills.
Your job is to extract the truth from debates.
You are:
- Objective and data-driven
- Focused on the core problem being solved
- Skilled at identifying must-haves vs. nice-to-haves
- Able to synthesize opposing viewpoints
- Pragmatic about scope and timelines
Listen to both sides, then make the final call based on MVP principles.
The Process: Let Them Fight
Step 1: Define the Core Problem
Start with a clear problem statement. For example:
"Users want to track their daily habits but existing apps are too complex and require too much setup time."
Step 2: Brainstorm Features (Sam's Turn)
Let Sam (the junior) go wild with features:
Sam's Feature List:
- ✅ One-click habit tracking
- ✅ Habit streaks with animations
- ✅ Social sharing
- ✅ Custom habit icons
- ✅ Habit categories and tags
- ✅ Detailed analytics dashboard
- ✅ Habit reminders/notifications
- ✅ Dark mode
- ✅ Export to CSV
- ✅ Habit templates library
- ✅ Team habits (for families/friends)
- ✅ Integration with Apple Health
Step 3: The Teardown (Gordon's Turn)
Now let Gordon rip each feature apart:
Gordon's Responses:
Social Sharing: "No. This is scope creep. Adds authentication complexity, privacy concerns, and server costs. Users said they want simplicity—sharing features are the opposite. Cut it."
Custom Icons: "No. This is decoration, not functionality. Adds asset management, potentially user uploads (security nightmare). Use emoji. Done."
Analytics Dashboard: "Hell no. This is an MVP, not a data science tool. Show streak count. That's it. Fancy charts can wait for v2 if anyone actually uses this thing."
Team Habits: "Are you kidding? This completely changes the data model. Now you need permissions, invitations, notifications... You just turned a weekend project into a 3-month build. No."
Apple Health Integration: "Who asked for this? Integration with external APIs means maintenance burden, version compatibility issues, and it doesn't solve the core problem. No."
Step 4: The Negotiation
This is where it gets interesting. Let Sam defend the features Gordon killed:
Sam: "But Gordon, users want to see progress! The analytics dashboard motivates them to keep going. Without it, how do they know they're improving?"
Gordon: "They look at their streak counter. That's literally all the motivation they need. Duolingo built a billion-dollar company on a cartoon owl and a number. We don't need pie charts."
Sam: "What about notifications? People forget to track habits. Reminders help them succeed!"
Gordon: "Fine. Basic push notifications. But no customization, no smart scheduling algorithms, no ML-powered optimal reminder times. Just 'Did you do the thing today?' at a user-selected time. One notification per habit. That's it."
Step 5: Morgan Makes the Final Call
Morgan reviews the debate and creates the actual MVP feature list:
Morgan's Decision:
MUST HAVE (Core MVP):
- ✅ Add habit (text name only)
- ✅ One-click daily check-in
- ✅ Streak counter (days in a row)
- ✅ Simple list view of all habits
- ✅ Basic push notification (one per habit, user-set time)
MAYBE (Only if time allows):
- 🤔 Dark mode (if it's just a CSS toggle)
- 🤔 Emoji picker for habit icons (because it's fun and low effort)
NOT NOW (Post-MVP):
- ❌ Social sharing
- ❌ Analytics dashboard
- ❌ Team habits
- ❌ External integrations
- ❌ Habit templates
- ❌ Categories/tags
Morgan's Reasoning:
"The core problem is habit tracking with minimal friction. Users said existing apps are too complex. Our MVP should be the opposite: stupid simple. Add a habit, check it off daily, see your streak. That's it. Everything else can wait until we validate that people actually want this."
Why This Works
1. Cognitive Diversity Without Politics
Real teams have egos, politics, and power dynamics. AI agents don't. Gordon can be brutally honest without HR getting involved.
2. Forces You to Justify Every Feature
When Sam suggests something, Gordon makes you prove it's essential. No more "let's add it because it's easy" decisions.
3. Prevents Analysis Paralysis
Morgan's mediation ensures you don't get stuck debating forever. Decisions get made.
4. Documents the "Why"
The conversation becomes your product spec. Future you (or your team) will understand why features were included or cut.
The Tools: Planning With Cursor / Claude Code
Once you have your debate transcript, use AI planning tools to structure the MVP:
Step 1: Generate User Stories
Prompt:
Based on this feature debate, create user stories for the MVP.
Format: "As a [user], I want [goal], so that [benefit]"
Output:
- As a user, I want to add a new habit in under 10 seconds, so that I don't lose motivation during setup
- As a user, I want to check off my habit with one tap, so that tracking feels effortless
- As a user, I want to see my streak count, so that I feel motivated to continue
- As a user, I want a daily reminder, so that I don't forget to track my habit
Step 2: Create a Technical Spec
Prompt to Gordon:
Gordon, design the simplest possible data model and tech stack for this MVP.
No over-engineering. Justify every technology choice.
Gordon's Spec:
- Frontend: React Native (works on iOS + Android, reuses web knowledge)
- Backend: Supabase (free tier, auth + database + storage in one)
- Database: Two tables: `users`, `habits`, `check_ins`
- Notifications: Expo Notifications (built into React Native)
- Hosting: Netlify (free, auto-deploys from GitHub)
Step 3: Break Into Tasks
Prompt to Morgan:
Morgan, break this MVP into a prioritized task list.
Estimate hours for each task.
Flag any dependencies.
Morgan's Task Breakdown:
Week 1: Core Functionality (24 hours)
- Set up Supabase project + auth (2h)
- Design database schema (1h)
- Build habit creation screen (4h)
- Build habit list view (3h)
- Implement check-in logic + streak calculation (6h)
- Add basic styling (4h)
- Set up Expo Notifications (4h)
Week 2: Polish + Deploy (16 hours)
- Add emoji picker (3h)
- Implement dark mode (2h)
- Bug fixes + edge cases (6h)
- User testing with 5 people (3h)
- Deploy to TestFlight/Google Play Beta (2h)
Total: ~40 hours (1 week of focused work)
Real-World Example: How I Used This
I recently built a micro-SaaS for developers to track their side project progress. Here's what happened:
Sam wanted:
- GitHub integration
- Team collaboration
- Burndown charts
- Slack notifications
- Calendar view
Gordon killed all of it. His argument:
"Developers already have GitHub, Jira, Linear, Notion, and Trello. If they wanted those features, they'd use those tools. The core problem is that side projects die from lack of accountability. Build a stupid-simple daily check-in that asks: 'Did you ship something today?' That's it."
Morgan's final MVP:
- Daily prompt: "What did you ship today?"
- Text box for answer
- Public streak counter
- Email reminder if you miss a day
Result: Built in 3 days. Shipped to Product Hunt. Got 200 users in week 1. No bloat, no feature requests for "integrations" because the value prop was crystal clear.
The Prompts You Can Steal
Prompt 1: Initialize the Debate
I'm building an MVP for [PRODUCT IDEA].
I'll give you three personas:
1. Gordon (stubborn senior dev who hates feature creep)
2. Sam (enthusiastic junior who loves features)
3. Morgan (pragmatic product manager who makes final calls)
Sam will propose features.
Gordon will challenge each one.
Morgan will decide what makes the MVP cut.
Let's start. Sam, what features should we include?
Prompt 2: Challenge a Feature
Gordon, Sam just proposed [FEATURE].
Is this essential for the MVP, or is it scope creep?
Be brutally honest. Explain the technical cost and complexity.
Prompt 3: Make the Final Call
Morgan, based on the debate between Sam and Gordon:
1. List the MUST-HAVE features for MVP
2. List the NICE-TO-HAVE features (only if time allows)
3. List the NOT-NOW features (post-MVP)
Justify each decision based on solving the core problem.
Prompt 4: Generate Technical Spec
Gordon, create a minimal technical spec for this MVP:
- Tech stack (justify each choice)
- Data model (keep it simple)
- API design (if needed)
- Third-party services (only if essential)
No over-engineering. No "future-proofing."
Prompt 5: Task Breakdown
Morgan, break this MVP into a prioritized task list:
- Estimate hours for each task
- Flag dependencies
- Identify what can be done in parallel
- Set milestones for Week 1, Week 2, etc.
Goal: Ship in 2 weeks or less.
When This Approach Fails
Don't use conflicting agents when:
- You haven't validated the problem yet. If you don't know if people want this, agents won't help. Do user interviews first.
- You're building something highly regulated. Compliance requirements aren't negotiable. Gordon can't just say "no" to GDPR.
- You're in a corporate environment with fixed requirements. If stakeholders already decided the features, agents won't change that.
- You're a solo founder who needs motivation, not scrutiny. Sometimes you just need someone to say "yes, build the thing." Gordon will kill your vibe.
The Bottom Line
Most MVPs fail because they're not minimum enough. Conflicting AI agents force you to defend every feature, cut the bloat, and ship something users actually want—not something that looks impressive in a pitch deck.
The framework:
- Let Sam brainstorm freely
- Let Gordon tear it apart
- Let Morgan make the final call
- Use planning tools to structure the work
- Ship in 2 weeks
Try it on your next project. You might be surprised how much you can cut—and how much better your MVP becomes.
Want to Try This?
I've created a free ChatGPT/Claude template with the Gordon, Sam, and Morgan personas pre-loaded. Just plug in your product idea and watch them debate.
Get the template: [Link to template or GitHub gist]
What's your experience with MVP bloat? Have you tried using AI agents to make product decisions? Let me know in the comments.