Directory Engine (Afterthedemo)

Leonard-powered niche directory of real shipped AI products (inspired by Startup Ideas Podcast + full Billy Howell episode).

A directory is curated data: “here are the best / most real X in this topic.” Built for AI agents. Research agent produces high-quality Markdown pages; Leonard (engineer) publishes. Later: affiliate collection, outbound deals, niche newsletter on a subdomain.

Target: $10K–$100K/month once traffic compounds. SEO and AI-answer engines increasingly favor curated sources.

Current Vertical: Afterthedemo

Real products that people actually shipped using AI tools (especially Hermes-style agent stacks). Only past-demo work with public, verifiable artifacts. No pure tool lists, no tutorial-first content, no client-service work.

See decisions.md and status.md for the locked filters and staged content pillars.

Core Playbook

Pick / lock the niche (taste + evidence bar).
Site foundation on Astro (Cloudflare + Porkbun + GitHub stack).
Research agent investigates one brand/product per cycle → polished Markdown page.
Engineer (Leonard) validates and commits.
Rinse until thousands of monthly visits.
Only after traffic: referral/outbound agent + newsletter cross-sell.

Rule: Never solicit deals before traffic exists.

Operating System (from full episode)

See best-practices.md for the complete system:

4-week cadence: Build team → Execute (no tinkering) → Hire/fire → Automate
Chief of Staff pattern (agents/chief-of-staff-prompt.md)
5-line routines
Adversarial QA loops
Anti-patterns to kill
Three proven shapes: Newsletter, Shopify, Directory/Site-builder

Repo Layout

directory-engine/
├── README.md
├── best-practices.md          # Full Grokbot operating system
├── decisions.md
├── status.md
├── agents/
│   ├── chief-of-staff-prompt.md
│   ├── research-prompt.md
│   ├── engineer-prompt.md
│   └── outbound-prompt.md     # future
├── templates/
│   └── page.md
├── pages/
└── src/                       # future Astro

Leonard Rules

Read status.md + decisions.md + best-practices.md first.
Small team only. Mission-critical agents.
Human decides stack and hard calls.
Quality and evidence over volume.
Automate only after the process is proven.

2.6 KiB Raw Blame History Unescape Escape