The $1M+ Solo AI Agent Business (Full Course)

Overview

Nick from Orgo walks Greg Isenberg through a complete playbook for running a solopreneur AI agent business charging $5K/month per customer with unlimited agents, usage, and support. The stack centers on Hermes agents running on Orgo cloud computers, Composio for tool connections, Agent Mail for agent emails, Obsidian as a context layer, and GPT-5.5 as the primary model, with Codex/Claude Code used to build agents for agents.

Key takeaways

Selling 'unlimited agents/usage' at $5K/month flat is economically viable because customers who think they need 100 agents actually only need 1-3 well-configured ones — the word 'tokens' should never appear in the offer.
The optimal target customers are legacy industries (law firms, insurance, marketing agencies, wholesalers, real estate) — not healthcare or finance — because they aspire to be AI-native but lack the in-house expertise to set up Hermes, Composio, or Obsidian themselves.
Hermes commands roughly 2x the price of OpenClaw ($10K vs $5K/month) because it's self-evolving and more reliable, and OpenClaw is already commoditized in late 2025.
Obsidian functions as a true second brain when fed daily transcripts (Nick uses a Limitless microphone) and structured markdown about people/projects — giving agents persistent context that simulates personal AGI.
The fulfillment scaling trick is using agents to build agents: Nick sends Telegram messages to a Hermes agent that uses the Orgo MCP to provision new customer VMs and install Hermes inside them, all while he's on a walk.
Composio solves the single biggest time sink in agent setup — authentication — by providing one MCP connector with managed auth across Gmail, Slack, GitHub, Notion, etc., eliminating insecure credential email exchanges.
Reliability requires a watchdog layer that auto-restarts crashed gateways plus an alert layer where each agent emails the operator from its own Agent Mail address when cron jobs or skills fail.

Chapters

0:00 The $5K/Month Solo Agent Business Premise

Charge $5,000/month per customer for managed AI employees
Customer never touches tokens, models, or infrastructure
Nick from Orgo will share offer, fulfillment, stack, and onboarding playbook

2:33 Crafting an Unlimited Offer Without Going Broke

Sell unlimited agents/usage/monitoring/support at flat $5K/month
Customers think they need 5-100 agents but really need 1-3
Avoid mentioning tokens or credits — it ruins the magic and adds friction
Position as selling an AI employee, not an AI agent

6:51 Choosing Verticals: Avoid Healthcare, Target Legacy Industries

Skip healthcare/finance due to regulatory burden
Target marketing agencies, law firms, insurance, manufacturers, wholesalers, real estate
These legacy industries want to be 'full-stack AI' but lack expertise
Niche down by geography or sub-vertical (e.g., commercial real estate in Florida)

12:00 Executive Pain Points That Generalize Across Industries

Decision-makers all suffer from too many emails, meetings, follow-ups, open loops
Build a base template for executive problems, then layer vertical skills
Example: matrimonial law firm gets case management + demand letter generation on top

14:30 Customer Acquisition Through Content and Async Fulfillment

Content is overpowered in 2026 — warm leads beat cold calls
Greg discovered Nick via late-night Instagram doom-scrolling about OpenClaw
Nick sends long-horizon tasks to his Hermes agent via Telegram while walking
Combine content reach with agent-driven fulfillment for maximum leverage

17:56 Customer-Facing Tooling Stack

Granola for meeting notes (has MCP) auto-syncs to Trello requests
Trello kanban limits scope creep to 1-2 requests per 48 hours
Loom for async progress updates, Superhuman for email velocity, Asana internal-only

20:49 Agent Stack: Hermes, Orgo, Composio, Agent Mail, Obsidian

Use Codex or Claude Code to build the agents you sell
Hermes preferred over OpenClaw — more reliable, self-evolving, can charge $10K vs $5K
Composio provides one MCP connector to thousands of apps with auth handled
Agent Mail gives each agent (e.g., 'Mia') its own email address
Obsidian vault since Nov 2025 acts as second brain feeding agent context

27:51 Model Selection and Why Cloud Computers Beat Mac Minis

GPT-5.5 is most token-efficient; GLM 5.1 and Kimi for cheaper open-source tasks; Opus 4.7 for long coding
Orgo gives each agent an isolated cloud computer it can operate visually
Mac minis create on-site debugging nightmares and lack sandbox safety
One workspace per customer, multiple agents per workspace, all manageable via Orgo MCP

36:00 Agents Building Agents: MCPs, Watchdogs, and Reliability

Nick installs Hermes into a new VM by telling his Telegram agent to use Orgo MCP
Critical setup MCPs: Perplexity, Exa, Context7, X MCP, Firecrawl — spawn sub-agents per source
Set up watchdogs to auto-restore crashed Telegram/WhatsApp gateways
Agents email Nick when cron jobs or skills fail so he fixes issues before customers notice

Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Greg Isenberg.