Official Lean AI Company Playbook
Step-by-step guide and everything you need to scale using AI while staying lean
Last week, I finished what might have been the most interesting research project of my life.
For the last three months, I've been obsessed with finding out how some companies are generating millions in revenue with teams so small that it should be practically impossible.
After having countless conversations with 50+ founders who are actually pulling this off, I wrapped it all up in one place.
I am happy to present the complete playbook these companies use to build, scale, and dominate their markets with skeleton crews.
Now, you might wonder why I'm sharing this instead of hoarding these insights. Fair question.
Here’s the short answer:
I exited my $150M+ ARR startup earlier this year. Made my money. Got my freedom. Now I want to do something useful with what I've learned.
There are millions of founders grinding away, working 80-hour weeks, raising round after round, slowly losing their souls to build companies the old way.
I've been there. It's exhausting and unnecessary.
These lean AI companies have found a better path. One that preserves your sanity, your equity, and your bank account.
Everything you are about to read comes directly from the founders who are living it.
Ready? Let's dive in.
Their operational workflows:
After 3 months of pestering founders, sitting through demos, and diving into their actual operations, here's what I discovered:
These companies have built entire operational systems that multiply human capability.
Customer Support:
Haven AI's bots handle 50% of tickets autonomously
Multiple companies process dozens of user interviews in a day (yes, one day)
Stan generates 100% of creator products from social media in minutes
ArcAds' support agent checks the knowledge base, responds automatically, and escalates only 20% of tickets
AI Dungeon's UGC Content Reviewer reduces misrated content while increasing moderation coverage
Finance:
Haven auto-categorizes over 80% of transactions
They reconcile 90%+ of accounts across thousands of customers
When one client buys Hubspot, the system learns and categorizes it for every future client automatically
Their prepaid expense detection proactively flags potential issues before accountants see them
ArcAds' AI Accountant pulls receipts from Gmail → categorizes with Claude → forwards everything clean to finance
Marketing and Sales:
Open Art cranks out hundreds of web pages weekly
One part-time person manages hundreds of SEM campaigns using OpenArt for creatives + Deepseek for copy
ArcAds' spy agent scrapes competitor ad libraries → feeds best into o1 → auto-generates viral videos
Their ghostwriter scrapes their own top-performing tweets → suggests daily ideas → sends to Notion
Development:
Every founder mentioned Cursor tripling their coding speed
Open Art supports millions of monthly active users with a tiny engineering team
Many of these companies run production without QA teams using Checkly for backend + Stably for end-to-end testing
Stan built entire platform with 3 founding full-stack engineers
ArcAds generates new keywords → creates landing pages with Claude → auto-adjusts bids daily
AI Dungeon's Bug Capture and Triage Agent monitors social channels for new issues and reduces the bug capture process time.
Their QA Agent analyzes code changes and generates comprehensive test plans
Their Tech Stack:
Here's where founders got surprisingly candid. Most wouldn't share workflows, but they named names:
Customer Support:
AI Support Agents: Systems that pull from knowledge bases to answer common questions
ZenDesk AI, ChatGPT API, custom knowledge base integrations
Email Management: AI-assisted drafting and response with human review
Feedback Analysis: NLP systems for synthesizing customer insights at scale
Claude for summarizing customer insights
User Research: Automated transcription, tagging, and insight generation
Dovetail for transcription and synthesis
Fathom for call analysis
Content Moderation: AI-powered systems for enforcing content policies
Cinder for content safety
Automated image and text screening
Finance & Operations:
Transaction Processing: Pattern recognition systems for categorization
Haven AI's pattern recognition systems
Document Processing: OCR tools for invoice and receipt extraction
OCR.space
Document AI
Reconciliation Systems: Auto-matching algorithms integrated with accounting software
Expense Management: Gmail receipt extraction
Claude for categorization
QuickBooks/Xero
Finance Tools:
Harvey for legal
Haven for accounting automation
Marketing & Content Creation:
Content Generation:
OpenArt for images/videos
ChatGPT for creative writing
Influencer Marketing:
Beacons AI for matching
Gmass for emails
Mightyscout for tracking
SEO Tools:
Ahrefs
SemRush with custom AI analysis layers
Content generation workflows for a company on the Lean AI leaderboard
Sales & Lead Generation:
Lead Enrichment: Automated data collection from multiple sources
Apollo[.]io
LinkedIn scrapers
Custom enrichment via Claude/GPT-4
Outreach Automation: Personalized communication at scale
Custom LinkedIn messaging via Claude
Sales Analytics: Revenue intelligence through automated data collection
HubSpot with AI layers
Custom dashboards with GPT-4 analysis
Meeting Tools: Transcription and actionable summary generation
Granola for meeting management
Fathom and GPT-4 for extracting key points
CRM Enhancement:
Attio (pulls emails/calendars automatically)
Development & Engineering:
Coding Assistance: AI pair programming tools that triple developer output
Cursor (3x developer output)
GitHub Copilot
WindSurf
Claude Code
Vector Databases: Semantic search and knowledge retrieval systems
Pinecone for semantic search
Weaviate
Chroma for embeddings storage
Event Processing: Event streaming and processing systems
Temporal.io
Apache Kafka or Confluent Cloud
Infrastructure Tools: Modern API and deployment infrastructure
FastAPI and Django for Python backends
Next.js API Routes for JavaScript backends
Vercel or Render for frontend
Admin Tools:
OnePassword
WhisperFlow for dictation
N8N for workflow automation
PandaDoc for contracts and signatures
Documentation: AI-generated technical documentation with human review
Where things break:
Around $2-3M ARR, many companies I spoke with hit the same wall.
Marcel from GrowthX showed me a $20,000 monthly bill from AirOps when their workflows became too complex.
That's when they realized they needed to build their own infrastructure.
The pattern repeats everywhere:
No-code tools work great for prototyping
They completely fail at scale
Workflows that took 30 minutes suddenly take 2+ hours
Simple changes require weeks of refactoring
Human errors multiply as people duplicate workflows for different clients
That's when companies start building custom solutions.
Expensive and time-consuming, but necessary.
A founder told me they didn't write a single line of code until they were almost at $2M ARR.
But at $3M, everything broke for them.
They built their own platform:
Reduced workflow execution from 2 hours to 2 minutes
Enabled real-time refactoring for new AI models (like GPT-4.1)
Created evaluation frameworks to measure AI output quality
Eliminated human errors in workflow duplication
Organizational design and hiring talent:
When it comes to running these lean AI companies, 90% of people assume these are lifestyle businesses run from Bali.
And trust me, nothing could be further from the truth.
"We are grinding hard at our offices," one founder told me bluntly.
It's just that these entrepreneurs have found a better way to build without giving away control and dilution. They work equally (if not harder) than you and I.
Core Hiring Principles:
Only technology-agnostic full-stack engineers
Generalists who think from first principles over specialists
People who can rapidly adapt to new AI tools
Fewer, better people with massive ownership
As Stan's Vitalii Dodonov put it: "When you hire the best people, they'll leverage AI to excel at their jobs. You don't need a hundred. Ten is enough."
The Global Talent Strategy:
Remote-first hiring expands the talent pool dramatically
International team members cost 50-70% less than US equivalents
Transparent compensation with growth paths regardless of location
Bar is extremely high: 7-step interview process, including assignments and reference checks
Team Structure Evolution:
As these companies scale, they organize differently:
Pod structure: Small, cross-functional teams with end-to-end ownership
Managing director model: Leaders who manage teams, bring in business, and develop practices
Engineering focus: Technical talent makes up a higher percentage of the headcount
Adaptation Speed Advantage:
The ability to quickly adopt new AI capabilities becomes a competitive edge:
Prototyping new solutions within days of new model releases
Refactoring workflows for newer, better models in hours instead of weeks
Continuous evolution of systems to improve accuracy and capabilities
Implementation best practices
1. Start with no-code, plan for custom
Use tools like N8N or AirOps for prototyping (all companies did this)
Plan for custom development at $2-3M ARR when things break
Document workflows as you build for easier migration
2. Measure everything obsessively
Track automation percentages by department
Monitor time saved vs. time invested
Calculate ROI on each automated workflow
3. Build for scale from day one
Design systems assuming 10x growth
Create modular workflows (easy to modify)
Plan for multiple models/APIs (avoid vendor lock-in)
4. Human-in-the-loop architecture
Never fully automate high-stakes decisions
Build confidence scoring for all AI decisions
Create easy override mechanisms
5. Documentation is everything
Document every workflow in detail
Create visual process maps
Record decision logic for AI systems
6. Build incrementally with regular feedback
Don't attempt to build end-to-end systems immediately
Implement in small chunks with frequent evaluation
Set up clear feedback mechanisms to improve performance
Create dashboards to track automation percentages and accuracy
7. Focus on data quality and context
Clean, structured data is essential for effective AI systems
Invest time in organizing existing information
Create clear documentation for internal processes
Build comprehensive knowledge bases before automation
8. Maintain human oversight and expertise
AI systems work best with human guidance and expertise
Create clear escalation paths for complex or uncertain cases
Design systems that learn from human corrections
Focus human time on high-judgment, strategic tasks
Your 6-month lean AI plan:
Unlike generic advice to "use AI," this is a specific, staged implementation plan that has worked across dozens of companies.
Follow it step-by-step:
Phase 1: Assessment and discovery (Weeks 1-2)
(I'm also opening a limited number of high-impact advising slots for serious founders and operators who want to work directly with me—beyond just consuming my content. Email or DM me)
Week 1: Operational workflow mapping
Objective: Identify and prioritize workflows for automation to maximize impact.
Why use the jobs-to-be-done framework?
The Jobs-To-Be-Done (JTBD) framework focuses on understanding customer goals, ensuring automation targets high-impact tasks. It shifts focus from product features to outcomes.
For example, Haven AI used JTBD to map invoice processing, identifying it as a 2-3 hour manual task ripe for automation.
Steps for workflow mapping:
Identify key jobs: List primary jobs per department:
Customer Support: Resolve inquiries quickly and accurately
Finance: Categorize transactions for accurate records
Marketing: Generate engaging content
Break down jobs into tasks: Decompose jobs into tasks. For support:
Receive inquiry → Categorize → Assign → Resolve → Follow up.
Document workflows: Record for each task:
Responsible party (e.g., support manager).
Time taken (use Toggl).
Tools used (e.g., email, CRM).
Pain points (e.g., manual categorization).
Identify pain points: Highlight slow, error-prone, or repetitive tasks.
Prioritize workflows: Use (Impact × Repeatability) ÷ Implementation Complexity.
Example: Automating support ticket categorization scores high due to frequent repetition and low complexity.
Tools and costs:
Metrics to track:
Workflows documented: Target 20-50.
Average process time: E.g., 2 hours for invoice processing.
Prioritization scores: Top 5 workflows ranked.
Week 2: Technical capability assessment
Objective: Evaluate tools and team readiness for AI adoption.