AI Stack Report: 5 companies scaling with fewer than 50 people
What tools, workflows, and decisions do the most efficient AI-native companies actually use? This report maps the stack behind Midjourney ($200M, 11 people), Cursor ($100M+, ~20 people), ElevenLabs ($500M, ~40 people), Bolt.new ($20M, ~30 people), and Lovable ($400M, 146 people). Free. No catch. Just the architecture.
Get the full report
Includes: detailed stack breakdown, workflow maps, key decisions, and what NOT to copy. Free. Instant download.
Also subscribe to Post AI Weekly (every Wednesday, free, no spam).
1. Midjourney — $200M with 11 people
The stack
| Function | Tool | Why |
|---|---|---|
| Distribution | Discord | 175M users already there. Zero acquisition cost. |
| Payments | Stripe | Direct subscription billing. No sales team needed. |
| Compute | Custom GPU cluster | Image generation at scale. Core IP. |
| Internal docs | Slab | Lightweight knowledge base. Replaced Notion. |
| Communication | Discord (internal) | Same platform as users. Radical simplicity. |
Key decision
No CRM. No marketing automation. No analytics beyond Stripe + Discord. David Holz's insight: every tool you don't need is a team you don't need to hire. The product IS the growth engine.
2. Cursor — AI-native development at scale
The stack
| Function | Tool | Why |
|---|---|---|
| Payments | Stripe | Subscription + usage-based billing. |
| Hosting | Vercel | Frontend and API routes. Zero DevOps. |
| Project management | Linear | Speed-optimized. Built for engineering teams. |
| Docs | Notion | Public changelog + internal wiki. |
| Support | Intercom + AI | AI-first support with human escalation. |
Key decision
The product (an AI code editor) means every employee already lives in the tool they're building. Dogfooding at a level most companies can't replicate. The tool IS the culture.
3. ElevenLabs — $500M ARR, 50% B2B, no sales org
The stack
| Function | Tool | Why |
|---|---|---|
| Cloud | AWS | GPU instances for voice models. Global edge. |
| Payments | Stripe | Self-serve B2B. Enterprise via Stripe invoicing. |
| Support | Intercom | AI chatbot handles 80%+. Human for enterprise. |
| Model serving | Modal | Serverless GPU. Scale to zero when idle. |
| Analytics | Mixpanel + Metabase | Product analytics + internal dashboards. |
Key decision
B2B without salespeople. Enterprise customers onboard via Stripe self-serve. The product is so good it sells itself. Support is AI-first by design.
4. Bolt.new / StackBlitz — $2M per employee
The stack
| Function | Tool | Why |
|---|---|---|
| Hosting | Vercel | WebContainer runs in browser. Vercel for deployment. |
| Database | Supabase | Postgres with real-time. Open source. |
| Project mgmt | Linear | Engineering velocity tracking. |
| Payments | Stripe | Subscription tiers. Usage limits. |
| AI | Anthropic + custom | Claude for code gen. Custom models for WebContainer. |
Key decision
30 people. $20M ARR. Every dollar of revenue funds compute, not headcount. The product itself is AI-mediated development — they literally eat their own dogfood at the architectural level.
5. Lovable — $400M with AI handling lead gen
The stack
| Function | Tool | Why |
|---|---|---|
| Cloud | Google Cloud | AI/ML services. Global infrastructure. |
| Docs | Notion | Company-wide wiki. Public docs mirror. |
| Project mgmt | Linear | Engineering + product. Speed as default. |
| Support | Intercom AI | AI-native support. Human for complex cases. |
| CRM | Attio | AI-powered CRM. Minimal manual entry. |
Key decision
146 people for $400M ARR is still 3-5x more efficient than traditional SaaS. The key: AI handles lead gen, onboarding flows, and tier-1 support. Humans focus on edge cases and relationship building.
Patterns across all 5
Stripe is universal. Every company uses it. Self-serve payments eliminate the sales ops function entirely.
Linear or bust. Fast-moving teams standardize on Linear. Speed of project management matters when you have 30 people doing the work of 300.
Intercom with AI. Not just a support tool — the first line of customer interaction. AI handles 70-80% of queries.
No Salesforce. Not one of these companies uses a traditional CRM. When the product is self-serve, CRM is overhead.
No dedicated DevOps. Vercel, Modal, Supabase. Platform-as-a-service eliminates the need for infrastructure teams.
Enjoying this? Get Post AI Weekly.
Every Wednesday. One deep analysis. 5 minutes. Zero fluff.
Subscribe free →