1,160 AI Business Ideas.
One Conversation.

What happens when you give 35 AI agents 40 prompts and ask them to systematically find the autonomous business worth building? A single session produced 1,160 scored ideas across 24 seed themes — and a thesis about why the AI layer is never the moat.

Ideas Generated

AI Agents

Prompts

Hours to Build

Explore the Ideas → Interactive Explorer →

The Arc

How This Was Built

The Spark

Started with a single observation: an AI agent that finds homes installing pools and mails a personalized postcard. That agent contains an entire business playbook. A public data source, an automated pipeline, a specific paying customer, and a delivery mechanism that reaches them before they know they need it. What else does?

The Method

24 seed batches. Two scoring frameworks. One for autonomous revenue generation. One for personal delight and learning. Every idea scored against 9 dimensions by Claude Opus agents running in parallel. 30 top ideas then analyzed by Haiku for napkin math and market sizing.

The Finding

The highest-scoring ideas share one pattern: public data that incumbents don't aggregate, turned into a service with a clear customer. The AI layer isn't the moat. The data pipeline is. The compounding advantage comes from connecting multiple trivial data sources into a single intelligence layer.

Worth Talking About

Three Ideas That Tell the Story

These aren't necessarily the highest-scoring. They're the ones that best illustrate why this approach works.

Founder's Seed

RouteKill

Gathers traffic data, organizes resident documentation, and submits formal petitions to Google Maps and Waze to deprioritize residential streets from GPS routing.

Claude Haiku · Analyst Note Addresses genuine neighborhood pain (cut-through traffic); Waze has merchant tools for similar use. Requires active HOA champions but repeatable. Risk: Waze/Google may not honor deprioritization requests consistently.

Trigger: Traffic complaint Revenue: $2,500 + $500/mo Break-even: 10 neighborhoods

4.75

Loved

ATLRestaurantDeathClock

Survival probability scores for every Atlanta restaurant. Your favorite brunch spot has a 34% chance of surviving another year.

Claude Haiku · Analyst Note Addictive "Doomsday" metric drives sharing and virality. Restaurant owners pay to fight the clock; advertising is obvious layer. Risk: accuracy backlash if scoring feels arbitrary.

Trigger: Weekly health data Revenue: Consulting + ads Break-even: 80 contracts

4.85

Home Appraisal Appeal Automation

CFPB formalized ROV rights in 2024. No startup has automated this. The comp engine is identical to property tax appeals.

Claude Haiku · Analyst Note CFPB-formalized right (2024); no automated player exists. Clear pain point (appraisal delays block closings). Risk: mortgage services may compete with own ROV products.

Trigger: Low appraisal Revenue: $299-599/filing Break-even: 600 filings/yr

4.36

Methodology

Four Phases of Ideation

Each phase expanded the search space by introducing new seed themes, data sources, and mental models. The ideas got sharper as patterns emerged.

Phase 1 — Seeds & Inspiration

Batches 1-10 · 500 Ideas

Door-to-door sales playbook meets AI targeting. Home services, animal-kingdom biomimicry, real estate triggers, B2B professional, consumer lifestyle, weird contrarian.

10 batches 500 ideas

Best: Property Tax Appeal Automator

Phase 2 — Expanded Seeds

Batches 11-14 · 200 Ideas

Directories, new parents, oh-shit triggers, boomers, World Cup opportunities, HOA compliance. More targeted seeds with stronger customer signals.

4 batches 200 ideas

Best: Internet Outage Refund Auto-Negotiation

Phase 3 — Greg Isenberg + Crappy Apps

Batches 15-19 · 250 Ideas

Newsletter-native businesses, app rebuilds for frustrated users, geo-credit signals, commercial real estate intelligence.

5 batches 250 ideas

Best: RedditAlerts.pro

Phase 4 — Data Asymmetry Deep Dive

Batches 20-23 · 150 Ideas

Court filings, neighborhood defense, NIMBY intelligence, permit pre-flight. The thesis crystallized into specific infrastructure plays.

4 batches 150 ideas

Best: ZoneWatch

Evaluation

Two Scoring Frameworks

Every idea was scored twice, through different lenses. The first framework asks: Can this run itself? The second asks: Would I want to build this even for free?

Autonomy-First Framework

9 dimensions · Weighted score 0-5

"Build once, collect forever. Ideas scored for how much they run without human intervention."

Autonomy

28%

Market Size

15%

Speed to First $

15%

Build Complexity

13%

Revenue/Customer

12%

Execution Fit

Defensibility

Ethics

Geo Advantage

Delight Framework

6 dimensions · Weighted score 0-5

"The most sustainable side project is one you'd build even if it never made money. Revenue is a bonus."

Fun / Funny

25%

Weird / Surprising

20%

Story / Thought Lead

20%

Autonomous Output

15%

Learning Value

10%

Life Improvement

10%

Infrastructure

Build One Pipeline, Unlock Many Ideas

The highest-scoring ideas cluster around shared data infrastructure. Six pipelines unlock 49 ideas. The first pipeline you build determines which ideas become trivial to launch next.

Results

Top Ideas by Framework

The top 24 ideas from each scoring lens. Cards show a radar-chart fingerprint of each idea's score profile. Click any card for the full breakdown.

Cross-Framework Analysis

Autonomy vs. Delight

Plotting ideas on both axes reveals four quadrants. The Gold Zone (top-right) contains ideas that are both highly autonomous and genuinely interesting to build. The Cash Machine quadrant works but may bore you. The Passion Project quadrant captivates but requires babysitting.

Research

The Playbooks Behind the Ideas

Four research tracks that shaped how ideas were generated and evaluated. Each playbook represents a different lens on what makes an AI business worth building.

What Makes a Good OpenClaw Project

OpenClaw is an AI agent runtime built on Claude with 30+ integrated tools. The projects that work best share a common architecture:

The OpenClaw Pattern

Trigger — a detectable event in a data source (new filing, weather threshold, price change, calendar date)
Intelligence — Claude analyzes the event and generates a recommendation, draft, or classification
Action — an automated output (email, letter, API call, database entry, alert)
Loop — the system monitors for the next trigger and repeats indefinitely

Best Practices

Start with data: the best OpenClaw projects are built backwards from an available data source, not forward from a product idea
Prefer public data: government databases, APIs, and scrape-able portals create durable pipelines that don't require paid subscriptions
Design for monitoring: build dashboards or alert mechanisms so you know when the pipeline is working and when it breaks
Contingency over subscription: for consumer-facing products, charging a percentage of value delivered (tax appeal savings, claim recovered) has lower friction than monthly subscriptions
One agent, one job: resist the urge to build a Swiss Army knife. The best pipelines do one thing reliably.

The Archetypal OpenClaw Business

Public record is published → agent detects record matching criteria → agent generates personalized output → output is delivered automatically → customer pays upon result

What Makes a Good Paperclip Project

Paperclip is a multi-agent orchestration platform. Unlike single-agent automations, Paperclip projects involve coordinating multiple specialized agents toward a shared goal, with a CEO agent managing the workflow.

When Paperclip Adds Value Over a Single Agent

Tasks too long for a single context window
Research that benefits from parallel exploration (multiple agents investigating different angles simultaneously)
Workflows with distinct phases (research → draft → review → publish)
Projects where you want human review at specific checkpoints without being in the loop for every step

The Paperclip Project Pattern

CEO agent: coordinates and delegates, never does the work itself
Specialist agents: deep work in a defined scope (research, write, code, analyze)
Output: structured deliverable that can be reviewed and approved before action

Best Paperclip Projects

Ongoing research digests (weekly market intelligence, competitor monitoring)
Multi-step document production (proposal → review → revise → publish)
Portfolio management (running parallel automations and surfacing anomalies)
Idea generation and scoring (exactly this project)

What Doesn't Work Well in Paperclip

Simple single-step automations (just use an OpenClaw agent)
Real-time responses (latency is too high)
Tasks requiring tight feedback loops with external APIs

The Directory Playbook (Greg Isenberg Framework)

The core insight: AI has flipped the economics of directory businesses. What previously required a team to build and maintain can now be bootstrapped solo for under $500, maintained for under $50/month, and defended through data that aggregates automatically.

The 4-Phase Directory Lifecycle

Data acquisition: scrape authoritative public sources (licensing boards, court records, permit databases, professional registries) to seed the directory with verified entries
Value differentiation: add a layer no competitor has — verification status, complaint history, performance data, local intelligence
SEO capture: programmatic pages targeting high-intent searches ("best [niche] in [city]") — one page per listing, one page per category × city combination
Monetization: typically 4 streams — listing fees, lead gen, affiliate commissions, premium profiles

What Makes a Directory Defensible

Proprietary data (licensed, scraped, or contributed) that competitors can't easily replicate
Verification layer (state licensing APIs, complaint history, background check integration)
Community engagement (reviews, Q&A, practitioner-contributed content that improves SEO automatically)
Local specificity (being the #1 resource for a specific city beats being #50 nationally)

The Directory Idea Evaluation Checklist

Is there a high-intent search query this would rank for? (e.g., "best pediatric dentists Atlanta")
Is there a data source that lets you populate 50+ entries without manual work?
Can you add a verification or rating layer that Yelp/Google doesn't have?
Is there a B2B customer (the listed businesses themselves) who would pay for a premium profile?
Can this generate a lead-per-customer worth $50-500?

The Data Asymmetry Playbook

The single most consistent pattern across the highest-scoring ideas in this project: turning public data that incumbents don't aggregate into a service that a specific customer will pay for immediately.

The Public Data Landscape Most Businesses Ignore

Data Source	What It Reveals	Best Customer
County assessor records	Property overassessment	Homeowners, real estate investors
HMDA mortgage filings	New HELOC/purchase signals	Home services, moving companies
Georgia SOS dissolutions	Company closings	Recruiters, suppliers, competitors
Georgia Odyssey court filings	Divorces, foreclosures, evictions	Real estate, legal services
BZA/NPU meeting agendas	Zoning changes before public hearing	Homeowners, developers
NHTSA recall database	Vehicle recall triggers	Auto services, attorneys
Downdetector API	ISP outages in real time	ISP refund services
FDA MAUDE database	Medical device adverse events	Plaintiff attorneys, regulators
UCC financing statements	Business debt signals	Credit, M&A, suppliers
OSHA complaint filings	Unsafe worksites	Attorneys, unions, competitors

The Pattern for Building on Public Data

Find a data source that updates regularly and is machine-accessible
Identify the customer who benefits most from that data being current and organized
Build the pipeline first (can you get 100 clean records in an afternoon?)
Build the delivery layer second (alert, report, API, or dashboard)
Test demand before scaling (10 customers paying before building the full product)

Pattern Recognition

What the Best Ideas Have in Common

Across 1,160 ideas scored through two independent frameworks, the ideas that rank highest on both share a common structure: they turn public data that incumbents ignore into a service that a specific customer would pay for immediately. There is no complex AI research required. There is no need for proprietary training data. The edge comes from the pipeline, not the model.

The playbook is consistent: identify a public data source that updates regularly. Automate its ingestion. Find the customer who pays the most for intelligence derived from that data. Deliver it faster than they could assemble it themselves. Fulton County assessor data becomes an automated property tax appeal business. Georgia court filings become real-time DUI attorney leads. BZA meeting agendas become zoning change alerts for neighborhood associations. Reddit posts become demand-signal intelligence for founders.

In every case, the AI layer is not the moat. GPT-4, Claude, Gemini — any of them can do the natural language processing. The moat is the data pipeline: the automated scraper, the structured database, the alerting logic, the delivery mechanism. A competitor cannot replicate the pipeline without building the same infrastructure from scratch, and most won't bother because the individual data sources appear trivial in isolation. The compounding advantage comes from connecting multiple trivial data sources into a single intelligence layer.

This is the pattern that separates the top 5% of ideas from the other 95%. Not smarter AI. Smarter plumbing.

1,160 AI Business Ideas.One Conversation.