The $5K/Month Solo
AI Agent Playbook

The pitch is dead simple. The customer doesn't touch tokens. They don't touch models. They don't touch infrastructure or security or any of the plumbing. They just get a digital employee that knows their business and gets better every single week.

That's the reframe the entire business runs on. You are not selling AI. You are removing every piece of friction between a business owner and a working AI employee.

And here's why it sticks. Once a business owner relies on that agent, taking it away is painful. They become dependent fast. That dependency is what makes a $5,000 a month retainer feel cheap to them, not expensive.

One thing before we go deeper. Nick's whole point is that you're probably more qualified than you think. If you can set up Claude Code, install an agent, or wire up an MCP, that's a skill 99% of businesses do not have time to learn. That skill is what you're selling.

When you start a one-person agent business, the customer does not want to think about tokens, infrastructure, security, or what happens when something breaks. They just want it to work. So Nick's offer creates abundance on purpose:

• Unlimited agents
• Unlimited usage
• Unlimited monitoring, support, and security
• Ongoing changes and improvements

You might flinch at "unlimited." Here's the trick. The customer thinks they need five, ten, a hundred agents. In reality they need one, maybe two, maybe three. One or two well-built agents go an incredibly long way. So "unlimited" sounds generous, but your actual cost stays controlled because nobody needs the volume they imagine.

The other rule: never say "tokens." The moment you mention credits or usage-based pricing, the customer starts wondering "how many do I have left?" and the magic dies. One flat number. Seamless experience. That's it.

The price he runs: $5,000 a month. It's been working.

This is the line that does the heavy lifting. Same technology. Completely different price tag.

An "AI agent" sounds like software. An "AI employee" sounds like a hire a business would happily pay five grand a month for. So clarify that you are not a commodity. You're not reselling Claude Code or ChatGPT. You're delivering a vertically specific, industry-specific employee that already understands their world.

Two more rules from this part:

• Speed. It should take no longer than 48 hours to get the first agent up and running for a customer.
• Talk in business outcomes, not time saved. "How much revenue can this generate" beats "how many hours this saves." Everyone's numb to "saves you time" by now.

Avoid to start: healthcare and finance. The regulatory burden and red tape make them a bad first market. Go here instead, where legacy industries are full of people, full of waste to automate, and want to be AI-native but don't know how:

How to niche down. These categories are huge, so pick a category, then a subniche. By geography ("real estate agencies in Florida") or by type ("commercial real estate agencies"). Nick's contrarian note: you don't have to start hyper-niche. Try a few industries, see where the market pulls you, then go vertical. Diverge, then converge.

Who's actually buying. Almost always an executive. And when you strip away the industry, they all have the same problems: too many emails, too many meetings, too many follow-ups, too many open loops, context scattered across too many projects and people.

So build a template that solves these executive problems out of the box, then layer the vertical-specific skills on top. For a matrimonial law firm, the base agent follows up and tracks projects, and then it also manages cases and drafts demand letters. Base layer plus vertical layer. That's the productized service.

The ideal sales position is when someone jumps on a call already knowing who you are and what you sell, before you say a word. You never want to sell to a cold audience.

So in 2026, content is the move. It puts your offer in front of customers, and it also gets you known, gets you on podcasts, and helps you hire. (Greg and Nick literally met because Greg saw Nick's face explaining an agent while doom-scrolling Instagram at midnight.)

If you have zero case studies, starting free is sometimes worth it just to get proof and referrals. Then let content do the warming.

The leverage compounds. Post something that reaches a lot of people, and have an agent fulfilling the work in the background. Nick literally sends long-horizon tasks to his agent over Telegram while he's on a walk, and comes back to finished work for his customers.

These are the tools the customer sees and touches:

• Granola for meetings. It has an MCP, so the agent has context over everything said. Notes auto-sync into requests on Trello.
• Trello as the customer-facing kanban (Backlog, To-Do, Doing, Done). The customer drags requests into To-Do. Limit them to one or two requests per 48 hours to prevent scope creep.
• Loom for updates. Send one whenever you ship something new. Showing the work visually lights people up.
• Calendly plus a personal website. Even a basic funnel books calls when content drives the traffic.
• Superhuman for plowing through customer emails fast.
• Asana for your own internal tracking (not customer-facing).

If you don't know how to build an agent, don't panic. You use an agent to build the agent. That's the whole trick. Here's the stack Nick installs for every customer:

Composio, Agent Mail, and Obsidian are the three Nick says everyone needs, no matter the industry. The first solves security and integrations, the second adds a personal touch and your alert channel, the third is the memory that makes the agent feel like it actually understands the business.

If the whole list feels overwhelming, here's Nick's exact stack in one line each: Codex to build, Hermes to sell (it doesn't break and it's self-evolving), Orgo so the agent has a real computer, Composio, Agent Mail, and Obsidian for every agent, and GPT 5.5 to keep it easy.

Nick has been building his Obsidian vault since November 2025. It holds everything about his people, his projects, his life. He even pipes daily transcripts from a Limitless mic into it.

His claim: this is what a real second brain feels like, not the "look, some research notes" version people usually demo. With a vault like this, the agent gets context on whatever it needs for any task. It feels like it never forgets and actually understands you.

That's the whole game. An agent that understands the business and has perfect context over everything it does. Markdown files are how you give it that.

Why virtual computers instead of buying Mac Minis? Because you need to manage your customers' setups remotely. A Mac Mini means driving somewhere to debug hardware, or dealing with a bricked machine after a bad update. Cloud computers fix that:

• Isolation and safety. Each agent runs in its own sandboxed cloud computer. If something goes wrong, delete it and spin up a fresh one in under a second.
• Scale. One platform, one connector, your agent reaching all of them. Nick showed 27 cloud VMs running across his workspaces, all managed from one spot.
• Clean structure. One workspace per customer's business. Each of that customer's agents lives inside their workspace. You manage all of it from anywhere, including from Telegram.

It's also a great demo. Showing a customer a visual computer that an agent is actually clicking around in makes the whole thing real in a way that describing "an agent" never does.

For your agent to set up other agents well, it needs current, real context. Docs change constantly, so ground every setup in live information using a few MCPs: Perplexity for up-to-date knowledge, Exa AI for real-time web search, Context7 for current docs straight from a project's GitHub, and the X MCP to pull great setups people post on Twitter.

Nick's power move: tell Claude Code or Codex to spawn five sub-agents, one per source. They each research independently, report back to the main agent, and you get genuinely best-practice setup instructions. Context is everything.

Then make it reliable, because this is what separates a real business from a hobby. The gateways that connect agents to Telegram or WhatsApp sometimes crash. Two things to set up so you fix problems before the customer ever notices:

• A watchdog. Tell your agent to set up a watchdog that auto-restores the gateway whenever it crashes.
• Alerts. This is where Agent Mail pays off again. Have the agent email you (from Mia's own email) the moment a cron job breaks or a skill fails. You fix it quietly. The customer never has to think about it.

Yes, Claude Code keeps getting better at general tasks. But building a specific agent, tailored to a specific industry, person, and workflow, is exactly the work most businesses can't and won't do themselves. People consistently underestimate how much that's worth.

That's the entire business: one person, plus an agent that builds agents, fulfilling for other businesses. $5,000 a month, per client.

It's a genuinely great time to be a solopreneur doing this. The skill is in your hands already. The only question is whether you go build the first one.