The Beginning: Automating Everything
Like many developers, I've been obsessed with using AI to automate my workflows. I started small—using individual n8n automations for individual tasks. The more I used it, the more I wondered: "How can I bring all these workflows together?"
Then the question shifted. It wasn't just about efficiency anymore. It became: "If AI can do this, what's stopping it from doing my entire job?"
That thought kept nagging at me. I'm a product manager, and I spend a lot of time creating the same artifacts over and over: user stories, acceptance criteria, feature breakdowns, database schemas, wireframes. These are the things that bridge the gap between "we want to build X" and "here's what engineers need to code."
What if I could build something that maintains the core context of a product idea and uses it to generate all the related artifacts? Not just one or two, but everything a product manager would typically create—and do it in a way that keeps everything in sync.
The Accident That Changed Everything
I had this idea brewing for months, but I couldn't find the time to work on it. Between my day job and life's usual demands, it kept getting pushed to the back burner. I was frustrated. I knew this could be something, but I just couldn't dedicate the time.
Then I broke my finger.
It was a stupid accident, honestly. But it gave me something I hadn't had in a long time: time off work. 3 months to focus entirely on building.
So I did (Thanks to speech-to-text and some one-handed typing…)
Building VibeMap: The Architecture
I called it VibeMap—a platform that takes your raw product idea and generates a complete development blueprint. But I didn't want to build a monolithic system that would be hard to extend. Instead, I designed a multi-agent architecture where each artifact type has its own specialized subagent.
The Multi-Agent System
The core idea is simple: maintain a single source of truth (the project context) and have specialized agents generate different artifacts from that context. Each agent is optimized for its specific task:
- Features Subagent - Generates prioritized feature lists with dependencies
- User Stories Subagent - Creates detailed user stories with INVEST principles
- Acceptance Criteria Subagent - Generates scenario-based criteria (happy path, edge cases, failure states)
- Personas Subagent - Builds detailed user profiles with demographics, goals, and pain points
- Pages/Components Subagent - Maps out UI structure with navigation relationships
- Schema Subagent - Designs complete database schemas with relationships and constraints
- UI Subagent - Generates wireframes and UI specifications
- Pseudocode Subagent - Creates implementation scaffolding
- File Structure Subagent - Proposes production-ready application architecture
- Project Summary Subagent - Generates executive summaries and project overviews
Context Management: The Secret Sauce
The real challenge wasn't generating individual artifacts—it was keeping everything in sync. When you update a feature, the related user stories, acceptance criteria, and database schema should all reflect that change.
I built a Context Management Service that:
- Maintains a comprehensive project context (project metadata, features, personas, pages, schema, relationships)
- Uses RAG (Retrieval Augmented Generation) to find relevant context for each generation
- Implements intelligent context pruning to stay within token limits
- Tracks relationships between artifacts so changes cascade automatically
The Generation Orchestration Layer
On top of the individual subagents, I added a Generation Orchestration Subagent that:
- Coordinates multi-step generation processes
- Manages dependencies between generation types
- Optimizes generation order and batching
- Handles cross-generation context sharing
This architecture means I can easily add new artifact types in the future. Want to generate API documentation? Add an API docs subagent. Need test plans? Add a test planning subagent. The system is designed to scale. The beauty of this is that the project plan provides the base that everything else needs.
What VibeMap Generates Today
Right now, VibeMap can take a product description like:
"I want to build a mobile e-commerce app where users can browse products, add items to cart, make secure payments, and track their orders. Sellers should be able to list products, manage inventory, and view sales analytics."
And generate:
- User Personas - Detailed profiles of buyers, sellers, and admins
- Feature Breakdown - Organized features with priorities (e.g., "User Authentication", "Product Catalog", "Shopping Cart", "Payment Processing")
- User Stories - Complete stories like "As a buyer, I want to add items to my cart so I can purchase multiple products at once"
- Acceptance Criteria - Test-ready criteria with tags (frontend/backend/shared) and scenario types
- Pages & Components - Suggested routes, navigation flows, and reusable components
- Database Schema - Complete ERD with tables, relationships, and SQL scaffolding
- File Structure - Production-ready directory layout with pseudocode scaffolding
All of this happens in real-time with streaming, so you can watch the blueprint build as it generates.
The Personal Touch: Why This Matters
Three months later, my finger has finally healed. I'm getting ready to go back to work. But now I have something I didn't have before: a product that I'm genuinely excited to share with the world.
Building VibeMap taught me something important. The question isn't "Will AI replace me?" It's "How can I use AI to do the work that matters?"
The repetitive, time-consuming parts of product management—creating artifacts, maintaining consistency, ensuring nothing falls through the cracks—those can be automated. What can't be automated (at least not yet) is the strategic thinking, the stakeholder communication, the judgment calls, the vision.
VibeMap doesn't replace product managers. It frees them to focus on the parts of the job that require human insight. It also helps non-technical founders communicate ideas to teams or have a solid foundation for any vibe coding they want to do.
What I Learned
Context is everything. The hardest part wasn't generating artifacts—it was maintaining context across all of them. A change to a feature should ripple through user stories, acceptance criteria, and schema automatically.
Modularity enables growth. The multi-agent architecture means I can add new capabilities without rewriting the core system. Each subagent is a self-contained module that follows consistent patterns.
Real-time feedback matters. Watching the blueprint generate in real-time isn't just a nice-to-have—it helps users understand what's being created and catch issues early.
The future of PM work isn't replacement—it's augmentation. Tools like VibeMap handle the mechanical work so PMs can focus on strategy, communication, and decision-making.
Call for Alpha Testers
VibeMap is in alpha right now. The core generation works well, but I'm actively refining prompts, improving the editing experience, and adding more customization options.
I'm looking for early users who are willing to try it out and share honest feedback. Especially:
- Product managers who want to see if the generated artifacts match what you'd create manually
- Startup founders who need to quickly turn ideas into actionable plans
- Development teams who've felt the pain of vague requirements leading to rework
- Developers who do PM work (I know many of you wear multiple hats)
Try it out:
- Live demo: Vibemap Demo (no signup needed)
- Full access: vibemap.ai (happy to provide coupons for free access to premium models)
What I'd love feedback on:
- Does the generated blueprint match what you'd expect from a senior PM?
- Are the acceptance criteria detailed enough for your team?
- What's missing that would make this indispensable for your workflow?
- Any edge cases where the generation falls short?
- How can the multi-agent architecture be extended for your use case?
I'm particularly interested in hearing from teams that have tried other AI planning tools or have specific workflows I should support. Your feedback will directly shape what VibeMap becomes.
The Journey Continues
My finger is healed, but the building isn't done. I'm back to juggling work and side projects, but now I have something concrete to show for those three months. More importantly, I have a clearer vision of how AI can augment—not replace—the work we do.
If you're curious about the technical implementation, the multi-agent architecture, or just want to chat about the future of product management, I'd love to hear from you. And if you try VibeMap, let me know what you think.
The future of PM work is being written right now. I'm excited to be part of shaping it.
Links:
- Try the demo: Vibemap Demo
- Full access: vibemap.ai
- Demo Video: Features Creation
- Questions? Drop a comment below or reach out!
Top comments (0)