Startups.do

Generate, evaluate, and assemble AI service companies at industrial scale— with Devin and the .do platform

Imagine Devin NOT just building software, but building real businesses defined in this repo.

---

Startups.do turns Industries × Occupations into production-ready startups—complete with Functions, Workflows, Agents, and assets.

• Vision — Business-as-Code: define the company in code; build & run the rest with AI.
• Output — MDX at scale: startups/{name}.mdx sites with YAML frontmatter.
• Engine — AI workflows: research, score, design, and assemble startup artifacts.
• Automation — Devin-ready: assign implementation to Devin via API and auto-generate the business.
• Metrics — Built-in evals: market sizing, unit economics, KPI targets per startup.

---

Links:

• See ROADMAP: ./ROADMAP.md
• See TODO backlog: ./TODO.md

Core Concepts

• Functions: Code, Generative, Agentic, Human
• Workflows: Deterministic and AI-driven steps that compose Functions
• Agents: Orchestrate research, generation, scoring, and iteration
• Business-as-Code: Declarative (Vision/Mission/Purpose, OKRs) + Imperative (Workflows), combined as code

Architecture Overview

• Framework: Next.js (App Router)
• Content model: startups/{name}.mdx rendered at runtime using next-mdx-remote-client
• Frontmatter: YAML stores structured generated objects:
  • Lean Canvas
  • Vision/Mission/Purpose (VMV)
  • Objectives and Key Results (OKRs)
  • Hypotheses and Experiments
  • Variants
  • Name/Brand/Domain (including builder domains/subdomains/mcp)
  • StoryBrand
  • Branding (style/theme/fonts/color/logo)
  • Landing pages (template/def/types/mdxui)
  • Stripe pricing and checkout
  • PRD
  • MVP scope/plan
• Components (to be added in components/): Hero, Problem, Solution, Features, Steps
  • If a component or the MDX file does not exist, default props are generated from YAML frontmatter

Workflow

1. Enumerate Occupations + NAICS codes
2. Research jobs, functions, and workflows for agentic automation
3. Score feasibility:
   • Can it be done remotely by a human on a laptop?
   • Capability for current models
4. Generate artifacts:
   • Lean Canvas
   • Vision/Mission/Purpose
   • OKRs
   • Hypotheses and Experiments
   • Variants
   • Name/Brand/Domain
   • StoryBrand
   • Branding (style/theme/fonts/color/logo)
   • Landing pages (template/def/types/mdxui)
   • Stripe pricing and checkout pages
   • PRD
   • MVP scope/plan
5. Assign to Devin via Devin API to build functions/services
   • Code → Functions → Workflow → Service → Business
   • Generate decks/slides
6. Save outputs to YAML frontmatter; render the site from MDX

MDX Format

• Location: startups/{name}.mdx
• Rendering: loaded at runtime via next-mdx-remote-client
• Default export: a page composed from components (Hero/Problem/Solution/Features/Steps)
• Frontmatter: YAML schema holds all structured business artifacts

Frontmatter Schema (excerpt)

---
name: Acme Auto Claims AI
slug: acme-claims
domain: acmeclaims.ai
score:
  remote_on_laptop: 0.9
  model_capability: 0.8
  overall: 0.85
vmv:
  vision: "Zero-touch auto claims in minutes."
  mission: "Automate FNOL to payout with AI and human-in-the-loop."
  purpose: "Faster, fairer claim experiences."
leanCanvas:
  problem: ["Slow claims", "Fraud risk", "Manual reviews"]
  solution: ["End-to-end agentic workflow"]
  uniqueValueProp: "Claims in minutes, not weeks"
  unfairAdvantage: "Proprietary datasets + insurer integrations"
  customerSegments: ["Auto insurers", "TPAs"]
  channels: ["Broker partnerships", "Direct"]
  revenueStreams: ["Per-claim", "SaaS"]
  costStructure: ["LLM", "Infra", "Integrations"]
  keyMetrics: ["Time-to-payout", "Fraud rate", "CSAT"]
okrs:
  - objective: "Reduce average claim cycle time"
    keyResults:
      - metric: "Cycle time"
        target: "≤ 5 minutes"
storyBrand:
  character: "Claims leaders"
  problem: "Backlogs and leakage"
  guide: "Agentic copilot"
  plan: "Pilot → Integrate → Scale"
  callToAction: "Start a pilot"
  success: "Minutes to payout"
  failure: "Status quo delays"
branding:
  nameIdeas: ["ClaimSpark", "Adjustly"]
  colors: { primary: "#0A84FF", secondary: "#111827" }
  fonts: { heading: "Geist", body: "Inter" }
  logoPrompt: "Simple spark + shield logo"
pricing:
  plans:
    - id: startup
      price: 199
      interval: monthly
      features: ["Up to 1k claims/mo"]
  stripe:
    productId: ""
    priceIds: {}
prd:
  summary: "Automate FNOL → triage → adjudication → payout"
  scopeMVP: ["Intake", "Doc parsing", "Risk scoring", "Human review"]
experiments:
  - hypothesis: "Automated triage reduces cycle time by 50%"
    metric: "Cycle time"
    variantCount: 3
variants:
  - name: "Self-serve insurers"
    changes: ["No human-in-loop"]
---

Components Contract

• Hero: { title, subtitle, ctaText, ctaHref }
• Problem: { bullets[] }
• Solution: { bullets[] }
• Features: { items[] }
• Steps: { steps[] }

If a component or props are missing, defaults derive from YAML frontmatter.

Integrations

• MDX loading: next-mdx-remote-client
• AI generation: Vercel AI SDK (@vercel/ai) to generate/transform objects, persisted to frontmatter
• Payments: Stripe pricing and checkout pages
• Automation: Devin API to build initial AI services / functions & MVP

Local Development

Install dependencies and run:

pnpm install
pnpm dev
# build/start/lint as needed
pnpm build
pnpm start
pnpm lint

This is a Next.js App Router project. Start with app/page.tsx or add MDX files in startups/ and components in components/ as you implement the roadmap.