March 15, 2025: First line of code ~8 weeks later: Working MVP 4 months total: First paying customer

Zero team. Zero additional funding. Just me, some AI coding tools, and a realization: eating your own dog food is the ultimate unfair advantage.

Here’s the reality: I built Zine in 8 weeks. But getting to market—positioning, onboarding, converting users—took another 2 months. Building is fast. Go-to-market takes time.

I’m going to tell you exactly how I did it, what worked, what didn’t, and why being a platform founder building on your own platform is different than you think.

The Backstory

For the past two years, I’ve been building Graphlit—a developer platform for AI-powered knowledge management. Think of it as the infrastructure layer that handles all the impossible parts:

• 30+ connectors (Slack, Gmail, GitHub, Notion, Linear, Jira...)
• Real-time sync with webhook management
• Audio/video transcription pipelines
• Vector embeddings + hybrid search
• Entity extraction (people, organizations, events, places)
• Multi-tenant architecture with data isolation
• Knowledge graph generation

We built this for developers who needed to add AI search and knowledge management to their products. Great product. Growing customer base. Classic B2B developer platform play.

Then I had an idea.

The Consumer Product Hidden Inside a Platform

“What if I used Graphlit to build a consumer product for myself?”

Not a demo. Not a reference implementation. A real SaaS that I’d want to use every day.

The idea: Search your entire team’s knowledge—Slack, email, meetings, GitHub, Notion—with AI. Semantic search. With context. With entities.

Traditional path: Build this from scratch

• 6-12 months
• 5-10 engineers
• $500K+ burn
• Custom auth for 30+ services
• Vector infrastructure
• Audio transcription pipeline
• Entity extraction at scale

My path: Use the platform I already built

• 8 weeks
• One person (me)
• AI coding tools
• Zero infrastructure work

Here’s the thing nobody talks about: if you’re a platform founder, you should be building products on your own platform. It’s the best way to understand what developers actually need.

The Stack (What Actually Shipped)

Frontend:

• Next.js 15 (App Router, RSC where it made sense)
• TypeScript (strict mode or go home)
• HeroUI (component library)
• Tailwind (no I’m not writing CSS in 2025)
• Vercel (push to deploy is still magic)

Backend:

• Next.js API routes (keeping it simple)
• Graphlit SDK (doing all the heavy lifting)
• Clerk (authentication)
• Redis via Upstash (caching)

AI Coding Tools:

• Claude Code (100% of development through September)
• Factory Droid (added later for system integration)
• Models: Claude Opus 4.1 → Claude Sonnet 4.5

What I DIDN’T build:

• Vector DB ❌
• Embedding pipeline ❌
• OAuth flows (mostly Arcade.dev) ❌
• Webhook management ❌
• Search infrastructure ❌
• Transcription service ❌
• Entity extraction ❌

Time saved: Roughly 4-5 months of pure infrastructure work.

Week 8: Working product. Not “infrastructure done, now we can start building.” Actual product ready for users.

What Made This Possible

1. Building on Your Own Platform

Here’s the unfair advantage: I had already spent two years building Graphlit. All the infrastructure I would have needed to build for Zine was already there:

• 30+ connectors (Slack, Gmail, GitHub, Notion, Linear, Jira, Google Drive...)
• Real-time sync across all sources (no webhooks to manage)
• OCR for PDFs, transcription for audio/video
• Vector embeddings and hybrid search (vector + keyword)
• Entity extraction (people, organizations, events, places)
• Knowledge graph automatically built
• Multi-tenant architecture with complete data isolation

What this meant: I could focus 100% on Zine’s UX and product experience. Zero time spent on infrastructure.

The lesson: If you’re building a platform, use it to build products. You’ll discover what’s missing, what’s broken, and what developers actually need. Eating your own dog food isn’t just good practice—it’s your competitive advantage.

2. AI Coding Is Real (But Not How You Think)

Here’s the uncomfortable truth: AI coding tools work. Really well. But only if you know what to ask for.

My setup:

Claude Code (March - September) = Everything From March through September, Claude Code handled 100% of development. Architecture, components, API integration, state management, everything.

“I need to add filtering to search results. Users should filter by source (Slack, Gmail, etc.), date range, and people. Show me the state management and Graphlit SDK integration.”

Claude gives me the full pattern. TypeScript types included. I tweak, ship.

Models: Started with Claude Opus 4.1 for complex architecture decisions, then switched to Sonnet 4.5 for speed once I had the patterns established.

Factory Droid (Added later) = System integration specialist After September, added Factory Droid for wiring complex systems together:

“Wire this Graphlit feed creation to the UI. Handle errors. Show progress.”

Understands both sides of the API boundary. Excellent for integration work.

The workflow:

1. I describe the feature (in product terms, not code)
2. Claude Code generates implementation
3. I review (looking for logic bugs, not syntax)
4. Test in browser
5. Ship

What changed: I went from writing code to reviewing code. That’s a 5-10x speed multiplier.

The catch: You need to know what good looks like. The AI won’t tell you if you’re building the wrong thing. That’s still your job.

3. An AI-Coding-Friendly SDK (That I Built)

One of the best decisions we made with Graphlit was building a TypeScript SDK with AI coding in mind:

• Clear, consistent patterns
• Comprehensive TypeScript types
• Great documentation
• Sensible defaults

When I used Claude to build Zine features, it could generate working Graphlit integration code because the SDK had obvious patterns. No weird edge cases. No arcane configuration.

The meta-lesson: When you’re building a platform, design for AI-assisted development from day one. Clear patterns and strong typing make the difference between “Claude can help” and “Claude can ship.”

What Zine Actually Does

Zine connects your team’s tools and makes everything searchable with AI:

Data sources: Slack, Gmail, Google Calendar, GitHub, Notion, Linear, Jira, Google Drive, and 20+ others.

Search: Semantic search across all sources. “What did Acme Corp say about pricing?” finds the email thread, Slack discussion, meeting transcript, and Linear ticket—in seconds.

AI Chat: Ask questions, get answers with citations. The AI has access to your entire team’s context.

MCP Server: Your IDE (Cursor, Windsurf, VS Code) can access your team’s knowledge via the Model Context Protocol.

Real-time sync: Connect once, syncs forever. New Slack messages appear in search immediately.

Start with Developers, Scale to the Business

Here’s something I didn’t expect: Zine started as a developer tool. I built it for myself—searching code discussions, PRs, architecture decisions.

Then I realized something: this problem exists for everyone.

Developers: Your coding agent searches Slack threads and meeting transcripts, not stale docs. MCP integration means your IDE has full team context.

Sales/CS: Never prep for a call again—search meeting recordings and Slack history with customers. “What did Acme Corp say about pricing?” One search, instant answer.

Product/Ops: Find any spec or decision—search Notion, Linear, and meeting notes. “Why did we choose approach X?” The full context is there.

Leadership: Daily audio briefs from all your tools. Know what’s happening without reading 200 Slack messages.

The insight: I built infrastructure for developers. But the product works for the whole company.

That’s the unfair advantage of building on your own platform—you can pivot from “developer tool” to “whole company tool” without rewriting infrastructure.

Technical Details (For the Curious)

Frontend: Next.js 15 + TypeScript + HeroUI + Tailwind CSS

Backend: Next.js API routes + Graphlit TypeScript SDK + Clerk

Caching: Redis via Upstash

Hosting: Vercel

Content/AI: Graphlit (our platform) handles all of it

What I didn’t build for Zine:

• Vector databases (already built for Graphlit)
• Embedding pipelines (already built for Graphlit)
• Search infrastructure (already built for Graphlit)
• 30+ connectors (already built for Graphlit)
• Webhook management (already built for Graphlit)
• Audio transcription (already built for Graphlit)
• Entity extraction (already built for Graphlit)
• Knowledge graphs (already built for Graphlit)

I had already built all of this for Graphlit over the past two years.

What Zine Actually Is (The Product)

Okay, enough about the tech. What does this thing do?

Core loop:

1. Connect your tools (Slack, Gmail, GitHub, Notion, Linear, Jira, etc.)
2. Zine syncs everything in real-time
3. Search semantically: “What did Acme Corp say about pricing?”
4. Get results across ALL sources: email thread + Slack discussion + meeting transcript + Linear ticket
5. Ask AI questions with full context

The killer feature: Entity extraction.

Zine knows “Acme Corp” is an organization. It knows “Sarah” is a person. It knows they’re related. So when you search “Acme Corp pricing”, you get every interaction with every person from Acme Corp, across every tool, with context.

Traditional search: keyword matching, 0 context Vector search: semantic but no structure

Zine: semantic + entities + relationships

Also shipped: MCP server. So your IDE (Cursor, Windsurf, VS Code) can access your team’s knowledge.

The whole thing is 100% real-time. New Slack message? Searchable in seconds.

The Uncomfortable Truths

1. You’re probably over-engineering

Look, I see builders spending 3 months on “the perfect architecture.” Meanwhile, I shipped in 8 weeks using one platform and AI tools.

The reality: If you’re still stitching together 10 services in 2025, you’re choosing infrastructure work over product work. That’s fine if that’s what you want to build, but be honest about it.

2. AI coding works (if you know what you’re reviewing)

The meme is “AI writes bugs.” True. But so do humans.

The difference: AI writes code in 30 seconds. You review for logic bugs (not syntax), test, ship.

Net result: 5-10x faster. But only if you know what good code looks like.

If you’re junior, AI might not help yet. If you’re senior, it’s a superpower.

3. Solo founder >> small team (sometimes)

Controversial opinion: for certain products, one person with AI tools beats a 3-person team.

Why? No coordination overhead. No meetings. No “aligning on architecture.” Just build, test, ship.

Obviously this doesn’t work for everything. But for a lot more products than people think. Add a second person when you have PMF, not before.

4. Users don’t care about your stack

Week 2, I’m obsessing over TypeScript strictness and component architecture.

Week 8, users tell me: “The meeting transcription is wrong sometimes.”

Nobody cares that it’s Next.js 14 with RSC. They care that it works.

The Unfair Advantage

Traditional teams spend 6 months building infrastructure before they can even start on product features.

I spent 6 weeks building product features. Zero time on infrastructure.

But here’s the real unfair advantage: I had spent two years building that infrastructure as a platform. Now I’m using it to build products in weeks instead of months.

If you’re a platform founder, this is your superpower. Use it.

What’s Next for Zine

We’re adding:

• More connectors (Confluence, Salesforce, HubSpot)
• Team data sharing (add colleagues to share data in the same Zine organization)
• Video indexing, search and analytics
• Live augmentation (AI assistant that watches meetings and surfaces relevant context in real-time)

And because I built the infrastructure layer in Graphlit, I can focus on these Zine features instead of scaling databases or optimizing search indexes.

This is the virtuous cycle: improving Graphlit makes Zine better. Building Zine reveals what Graphlit needs. Both products win.

The 2025 Builder Playbook

If you’re building in 2025, here’s what I learned:

For platform founders:

• Build products on your own platform. Seriously.
• Your best customers are yourself
• The dogfooding will reveal every rough edge
• It’s the fastest way to find product-market fit for your platform

For product builders:

• Don’t build infrastructure if you can avoid it
• Find platforms that abstract the hard stuff
• Use AI coding tools (Claude Code, Factory Droid, whatever works)
• Ship in weeks, not months
• Stay solo until PMF

The formula for platform founders: Your platform + AI tools + your own needs = unfair advantage

The formula for everyone else: Someone’s platform + AI tools + good taste = competitive advantage

What This Means for You

Three years ago, you needed a team and 6 months to build Zine.

Today, you need 8 weeks, AI tools, and either your own platform or someone else’s.

The barrier isn’t technical anymore. It’s “do you know what to build?”

Product sense > engineering chops in 2025.

---

Try It

Zine: zine.ai (the product, free to start)

Graphlit: graphlit.com (the platform that made it possible)

Questions: @kirkmarple on X