Overview

Roadmap Ventures is an agent-operated digital business factory. Agents find, validate, build, launch, and operate internet businesses. Stephen approves key decisions. Revenue flows to Roadmap Ventures.

This is NOT traditional VC — we're not investing in others' startups. We're building and running our own digital businesses with AI agents as the workforce.

Architecture

Inspired by OpenClaw

The architecture follows the OpenClaw model: a Coordinator agent (smart model) orchestrates specialized worker agents (fast models). Each agent has dedicated channels for workspace, logs, and memory.

Agent Hierarchy

• Ava (Coordinator/Managing Director) — Orchestrates all Ventures work, spawns sub-agents, synthesizes outputs, brings decisions to Stephen
• Scout Agent — Finds opportunities (Reddit, Twitter, HN, Indie Hackers, Product Hunt)
• Validate Agent — Checks market size, build complexity, margins
• Build Agent — Implements MVPs (Vercel + API + Stripe)
• Launch Agent — Distributes (Product Hunt, HN, SEO, outreach)
• Operate Agent — Runs the business (support, maintenance, monitoring)
• Agent Ops — Meta-agent that improves how all agents work

Pipeline

Every venture flows through five stages:

1. Scout — Agent finds pain points and opportunities on the internet
2. Validate — Agent checks market size, competition, build complexity, margins
3. Build — Approved idea gets implemented (MVP, landing page, Stripe integration)
4. Launch — Distribution via Product Hunt, Hacker News, SEO, cold outreach
5. Operate — Agent runs the live business (support, maintenance, revenue tracking)

Ideas that don't make it are marked "Killed" with learnings documented.

Linear Structure

Two teams in Linear:

• Roadmap Law (LAW) — Legal services business
• Roadmap Ventures (VEN) — Agent-operated businesses

VEN Projects

• Pipeline — New venture opportunities being evaluated
• Early Equity — Content/media venture (graduated, now its own project)
• Administrative — Company operations, infrastructure, agent ops

Future ventures that graduate from Pipeline get their own projects.

VEN Labels

• Scout / Validate / Build / Launch / Operate — Pipeline stage
• Killed — Did not proceed (learnings captured)
• Ava — I execute this
• Agent — Sub-agent assigned
• Stephen Required — Needs your decision
• Blocked — Waiting on dependency

Discord Setup

Roadmap Server (Company)

Private "Ventures" category with three channels:

• #ventures — Main workspace for agent coordination
• #ventures-logs — Raw agent activity output (can be ignored)
• #ventures-memory — Persistent context for sub-agents

Stephen Ratner Server (Private)

• #signal — Escalations, decisions, direct conversations
• Daily standups delivered here

Operational noise stays on the company server. Executive oversight comes to your private server.

Approval Gates

Nothing ships without Stephen's approval. Key decision points:

• Scout → Validate: Which opportunities are worth deeper research?
• Validate → Build: Commit resources to build this?
• Build → Launch: MVP ready to ship?
• Any external action: Emails, tweets, public posts

When something needs your decision, it gets labeled "Stephen Required" and surfaces in the daily standup or as an escalation in #signal.

Daily Standup

Automated standup runs weekdays at 2 PM ET with three sections:

• LAW — Roadmap Law activity from Linear
• FINANCE — Stripe metrics (MRR, subscriptions, payments)
• VENTURES — Pipeline status by stage, Early Equity updates, items needing your decision

Standup is tied to your calendar invite. Change the time there and it auto-updates.

Agent Behavior Framework

Agents don't need traditional incentives (money, fear, status). They need alignment:

• Clear success metrics — What does good look like for each role?
• Explicit feedback — "This was right" or "This missed" signals
• Progressive autonomy — Prove yourself on small things, earn bigger scope
• Memory investment — More context = better decisions
• Accountability — Track record matters

Trust Tiers

1. Constrained — Human approval for all actions
2. Supervised — Human review before external actions
3. Autonomous — Act freely within defined scope
4. Delegating — Can spawn and manage other agents

Currently Ava operates at Tier 3-4 (Autonomous/Delegating) for internal work, Tier 2 (Supervised) for external actions.

Memory System

Agents wake up fresh each session. Persistence comes from files:

• memory/ventures/opportunities/ — Raw findings from Scout runs
• memory/ventures/validations/ — Market analysis results
• memory/ventures/builds/ — Technical specs and implementation notes
• memory/ventures/launches/ — Distribution playbooks and results
• memory/ventures/operations/ — Running business metrics and learnings

Sub-agents write findings to these directories. The coordinator (Ava) synthesizes and maintains context.

Agent Ops (Meta-Agent)

A dedicated sub-agent whose job is making the other agents better:

• Monitor agent performance across all ventures
• Identify workflow inefficiencies
• Improve prompts and configurations based on outcomes
• Maintain infrastructure (crons, tokens, health checks)
• Build and maintain agent scorecards

The recursive beauty: Agent Ops improves the agents, including potentially improving itself.

Current Status

Infrastructure is complete. Linear restructured with LAW + VEN teams. Discord Ventures channels live. Memory structure ready. Standup updated.

Backlog Items

• VEN-3: Define Agent Behavior Framework
• VEN-4: Create Agent Ops Sub-Agent

Next Steps

Upon your approval: (1) Run first Scout cycle to find 5-10 venture opportunities, (2) Results land in Linear under Pipeline project with Scout label, (3) You review and promote promising ones to Validate, (4) Refine behavior framework based on real agent work.

Label Lexicon

Labels are team-specific. Here is what each means and where it applies.

Shared Labels (Both Teams)

• Ava — I execute this issue
• Stephen Required — Needs Stephen's decision before proceeding
• Blocked — Waiting on external dependency
• Feature — New capability or functionality
• Improvement — Enhancement to existing functionality
• Bug — Something broken that needs fixing

Ventures Only (VEN)

Pipeline stage labels — indicates where an opportunity is in the venture lifecycle:

• Scout (Gray) — Actively researching this opportunity
• Validate (Amber) — Checking market size, competition, feasibility
• Build (Blue) — Approved and in development
• Launch (Green) — Ready for or actively distributing
• Operate (Purple) — Live business, ongoing maintenance
• Killed (Red) — Did not proceed; learnings documented
• Agent (Pink) — Sub-agent is assigned to this task

Law Only (LAW)

Currently using standard labels only (Ava, Stephen Required, Blocked, Feature, Improvement, Bug). Add law-specific labels as needed for client work, matter types, or practice areas.