Nick charges $5,000 a month. Per customer.
Not for software. Not for a course. For a digital employee that knows the customer's business and gets better every week.
He runs the whole thing alone. No team, no office, 27 customer machines humming in the cloud while he goes on walks and fires off tasks from Telegram.
This is the solo AI agent business, and Greg Isenberg just got Nick from Orgo to lay out the entire playbook on camera: the offer, the industries that actually pay, the stack, and how to fulfill without drowning.
I watched the whole thing so I could pull out the parts that matter. If you already know how to set up Claude Code or an agent harness, you're sitting on a skill that law firm partners will pay $60K a year for. Most founders don't realize this yet.
Here's the breakdown.
Watch the full interview first
The episode is a full course, not a teaser. Nick screen-shares his actual setup, his Telegram agent, and his customer workspaces.
Now the playbook, piece by piece.
The offer: unlimited everything at $5,000 a month
Nick's offer sounds insane on paper.
Unlimited agents. Unlimited usage. Unlimited monitoring, support, security, and ongoing changes. $5K a month, flat.
How is that not a money pit?
Because customers think they need 10 or 100 agents. In practice they need one, two, maybe three. One well-built agent with proper context does an absurd amount of work. So "unlimited" costs Nick almost nothing extra while removing every objection from the sales call.
The rules he follows on positioning:
- Never say the word "tokens". The customer doesn't touch models, credits, or infrastructure. The moment you mention usage limits, the magic dies and they start counting.
- Sell an AI employee, not an AI agent. The agent gets a name, an email address, a personality. "Mia" who handles your follow-ups beats "an automation workflow" every time.
- First agent live within 48 hours. Speed to value closes the deal and kills buyer's remorse.
- Talk business outcomes, not time saved. "Time saved" is worn out. Revenue generated is not.
One customer at $5K/month is $60K a year. Ten customers is $600K a year, solo, with agents doing the fulfillment. That's how you get to the $1M+ number in the title without hiring anyone.
Who actually buys: law firms, not startups
This part surprised me.
Nick doesn't sell to tech companies. They can build this themselves. He sells to what he calls legacy industries that want to be AI-native but have no idea how:
- Marketing agencies
- Law firms
- Insurance agencies
- Manufacturers and wholesalers
- Real estate agencies
He deliberately avoids healthcare and finance. Too much regulation, too much red tape for a solo operator.
And the buyer inside those companies is always the same person: the executive. Abstract away the industry and every decision maker has identical problems. Too many emails. Too many meetings. Too many follow-ups and open loops across too many projects.
So Nick builds a template that solves executive problems out of the box, then layers industry-specific skills on top. For a law firm: case management and demand letters. For an agency: reporting and client follow-ups.
Greg's addition here is smart: niche down after you start, not before. Try a law firm, try an agency, see where the market pulls you, then go vertical. "Commercial real estate agencies in Florida" is a wedge. "Wholesalers" is an ocean.
The stack Nick runs the business on
Customer-facing tools first:
- Granola for meeting notes, synced automatically into Trello
- Trello as the customer-facing kanban board. Customers drag requests into a to-do list, capped at one or two requests per 48 hours so scope creep doesn't eat him alive
- Loom for async update videos, sometimes recorded at 2am
- Calendly plus a personal site. His words: "I have a horrible funnel." It still books calls
- Superhuman for email, Asana for internal tracking
Then the agent stack:
- Hermes as the agent harness customers actually use. Nick's one-liner: it doesn't break and it's self-evolving. He'd rather sell Hermes at $10K/month than OpenClaw at $5K ("OpenClaw commoditized it already")
- Claude Code or Codex to build the agents. You don't sell these, you build with them
- Orgo for cloud computers the agents live in. One workspace per customer, isolated, deletable in a second. Nick is the founder, so yes, he's biased, but the logic holds: you can't debug a Mac mini sitting in a law firm's closet
- Composio for connections. One MCP that handles auth and tool calls for Gmail, Slack, Notion, thousands of apps. Nick says authentication is the single biggest time sink in this business, and this kills it
- AgentMail so every agent gets its own email address. Mia emails Nick when her cron job breaks, before the customer ever notices
- Obsidian as the memory layer. Nick has been building his vault since November 2025 and calls it the closest thing to personal AGI he's experienced
On models: GPT 5.5 for the agents (efficient with tool calls, doesn't eat tokens), GLM 5.1 or Kimi for cheap lightweight tasks, Opus 4.7 for long coding runs.
If the harness debate interests you, I wrote about where this whole category is heading in AI agents are the new SaaS.
Agents that set up other agents
This is the fulfillment unlock, and it's the reason one person can serve dozens of customers.
Nick doesn't manually configure anything. His own Hermes agent uses the Orgo MCP to spin up customer computers, install harnesses, and configure everything remotely. On the episode he asks it "how many Orgo VMs do I have?" from Telegram and it comes back: 27, all running, broken down by customer.
A customer emails that something broke? He forwards it to his agent from his phone and it goes and fixes it.
Two reliability tricks he insists on:
→ Watchdogs. Gateways (the bridge between the agent and Telegram or WhatsApp) crash. Tell your agent to set up a watchdog that auto-restores any crashed gateway before the customer notices.
→ Agents that report their own failures. Every customer agent emails Nick when a cron job or skill fails. Observability, but the agent is the one doing the observing.
He also feeds his build agents live context through the Perplexity, Exa, Context7, and X MCPs, so setup instructions come from today's docs, not stale training data.
This pattern (one orchestrator agent managing a fleet of worker agents) is the same one Andrew Wilkinson uses to run his life and portfolio. I broke that down in how Andrew Wilkinson uses AI agents.
How he gets customers with a "horrible funnel"
No cold calls. No ads. Content.
Greg found Nick doom-scrolling Instagram at midnight, saw him demo OpenClaw, and booked him on the podcast. That's the acquisition channel working exactly as designed: when a prospect gets on a call already knowing who you are and what you sell, the call is a formality.
Nick's take is that posting content is the single best use of a solo operator's time right now, because AI handles the fulfillment while the content compounds. His day: post, go on a walk, send a long task to his agent from Telegram, come back to finished work.
Early on, he says, work for free if you have to. Case studies and referrals are worth more than the first few checks.
What I'd steal from this playbook
Three things stand out to me after building with agents myself.
First, the abstraction insight is the whole business. Every executive has the same four problems (email, meetings, follow-ups, open loops). Build the template once, sell it many times, customize 20% per vertical. That's productized service economics with software margins.
Second, "unlimited" is pricing genius precisely because demand is imaginary. Customers buy the ceiling, use the floor. You'd never dare this with human labor.
Third, most businesses don't need a full digital employee on day one. They need their five most painful workflows automated, and that's often the wedge that gets you in the door. If you want to see what those look like in practice, this catalog of business automations is a solid map of what companies actually pay to fix first. Start there, then upsell the agent.
The window matters too. Nick's edge isn't secret tech, it's that 99% of businesses are behind on AI and he isn't. That gap closes. The founders who build the customer relationships now keep them when it does.
FAQ
How much can you charge for an AI agent service?
Nick charges $5,000 per month per customer for unlimited agents, support, and ongoing improvements. He jokes that Hermes-based agents could justify $10K/month. At 10 customers that's $600K a year for a one-person business, with agents handling most of the fulfillment.
Do I need to know how to code to start an AI agent business?
Less than you'd think. Nick's core method is using agents to build agents: Claude Code or Codex does the actual setup work, guided by up-to-date docs pulled through MCPs like Perplexity and Context7. What you need is enough fluency to direct the tools and debug when they stall.
What industries buy AI agents?
Marketing agencies, law firms, insurance agencies, manufacturers, wholesalers, and real estate agencies. They're people-heavy, they want to be AI-native, and they can't build it themselves. Nick avoids healthcare and finance because of regulatory burden.
What's the biggest risk in running this solo?
Fulfillment collapse. Customers become dependent on their agents fast, and when something breaks it's painful. Nick's answer: cap requests at one or two per 48 hours, set up watchdogs that auto-restore crashed gateways, and have every agent email you when its own jobs fail.
Steal playbooks like this every week
I interview bootstrapped founders doing $100K to $10M a year on the Profitable Founder Podcast, and I dig for exactly this kind of detail: real prices, real stacks, real customer counts.
