Inflation Shock, AI Employees Are Replacing Real Work

● Inflation Shock

The core of a solo AI agent business that earns $5,000 a month is not “AI agents,” but “AI employees.”

Customers do not care about tokens, models, or infrastructure.

What customers want is a “working system,” not an explanation of AI technology.

The real point of this interview is not simply how to build agents, but which industry to choose, what offer to sell, and what tool stack to operate in order to grow into a business that can earn $5,000 or more per month.

In particular, the points below are key.

Why people who are familiar with AI are the ones who can make money
Why a “unlimited” offer is actually a powerful sales strategy
Why legacy industries are a better starting point than healthcare or finance
Why content is stronger than cold calling for customer acquisition
Why the essence of an agent business is operational stability, not technology
Why tools like Obsidian, Composio, and Agent Mail are effectively essential
And why the “agents setting up agents” structure, which is rarely discussed elsewhere, is a game changer

Why solo AI agent businesses are taking off

People who know how to use AI are already a scarce resource

People who have used AI often underestimate the skills they already have.

But reality is different.

Most companies in the world are still slow to adopt AI, unable to set it up, and unable to operate it properly.

In other words, the ability to set up tools like Claude Code, Hermes Agent, OpenClaude, and Codex is already a monetizable skill.

This is not just a hobby, but a business capability that closes the productivity gap.

AI technology is evolving rapidly, while customer adoption is much slower.

That gap is exactly the business opportunity.

Now is the best time for solopreneurs

In the past, building automation systems required a development team, an operations team, and a security team.

Now one person can do it.

Because you can deploy not just one agent, but multiple agents, to handle repetitive work for another business.

In other words, the combination of “solo business owner + AI agents” becomes a small operating company.

Offer design determines everything

What you should sell is not AI, but an “employee”

This is the most important message.

Customers are not buying an “AI agent.”

Customers are buying an AI employee that works inside their company.

So it is best to avoid words like “tokens,” “models,” and “usage” as much as possible.

The moment those words appear, the customer’s mind switches into technology-buying mode.

Then emotion cools, and the magic is broken.

On the other hand, saying “this employee will organize your email, follow up, and track your tasks” is much easier to understand.

Why an unlimited offer works

The structure proposed in this case is very strong.

Unlimited agents
Unlimited usage
Unlimited monitoring
Unlimited support
Unlimited security response
Continuous improvement

This is charged at $5,000 per month.

At first glance, it looks risky, but in reality, the customer does not need nearly as many agents as they imagine.

Usually, one to three is enough.

In other words, “unlimited” is not a word that literally allows infinite usage, but rather a sales phrase that removes psychological barriers.

This is more important than it may seem.

When customers see usage limits, they immediately start calculating.

At that point, they are no longer buying; they are evaluating.

You must show the first agent within 48 hours

The offer must be fast.

If the first agent is not moving within 48 hours, it will not work.

The reason is simple.

While customers wait, they feel more anxiety than anticipation.

In other words, the speed of initial setup is trust.

Which industries are the best fit

Healthcare and finance are not recommended at first

Highly regulated industries may look attractive, but they are too burdensome as a starting point.

Licensing, security, liability, and compliance are complicated.

For a solo business owner in the beginning, it is too heavy.

Legacy industries are the opportunity instead

The recommended industries are:

Marketing agencies
Law firms
Insurance agencies
Manufacturers
Wholesalers
Real estate brokerages

The common denominator is clear.

They may not be the most technologically sophisticated industries, but they have a very strong desire to grow with AI.

And internally, they have many problems to solve.

Too many emails
Too many meetings
Too many follow-ups
Too many open tasks
Too much context across people and projects

In the end, the industries differ, but the pain points are similar.

What matters here is not industry-specific features, but first solving the common problems decision-makers face, and then layering industry-specific features on top.

It is fine to start broad and narrow down later

Many people think they need to find a super niche from the very beginning.

But this interview takes a slightly different view.

At first, it is okay to test multiple industries.

What matters is not wandering for too long.

Try multiple industries
See market responses
Find the market that fits you best
Then move into a super-vertical from there

In other words, broaden first, then narrow.

Customer acquisition is stronger through content than cold calling

The ideal customer is someone who already knows you

Cold calling from scratch is inefficient.

The best lead in this business is someone who already knows who you are and what you do.

That is why content matters.

When people see your content, they already come in with some level of trust.

It converts much better than cold calling.

Free pilots are also a strategy in the beginning

At the start, you may also begin for free in order to build case studies.

That is not a loss; it is an investment.

Once you have real case studies, sales become much easier.

As of 2026, content-based trust building is especially powerful.

The customer communication tool stack determines real-world execution

Granola and Trello organize meetings and requests

The first step in customer communication is not losing the meeting context.

This is where Granola comes in.

Records meeting content
Organizes context
Makes it usable for later agents

And this note then connects with Trello.

Trello is not just a board; it is a scope management tool.

If requests pile up, the business breaks.

So you need to manage them as backlog, to do, in progress, and done, while handling customer requests in a limited and controlled way.

Loom, Calendly, Superhuman, and Asana make operations more robust

Loom: Quickly deliver progress updates via video
Calendly: Automate meeting scheduling
Superhuman: Fast email handling
Asana: Track internal details

The important thing is not using many tools.

Each tool must have a clearly defined role.

Customers do not want a complicated system.

But operators must absorb complexity.

The core of the agent-building stack is “agents building agents”

Claude Code and Codex are the foundation

Building agents directly is good, but in practice, having another agent build them is much more efficient.

Claude Code
OpenAI Codex

These two tools are used to build customer-facing agents.

In particular, desktop apps are a strong advantage.

Hermes Agent is more suitable for customers

Rather than selling Claude Code or Codex directly, Hermes Agent is more practical for real customers.

The reason is stability.

And because the harness structure is good, it is easy to switch models.

In other words, you are not tied to a specific model and can move to the latest one.

That is a surprisingly big advantage.

The AI model environment changes too quickly.

Why Orgo is more flexible than Hostinger

Hosting can be done anywhere, but Orgo is especially powerful.

Because:

You can manage multiple agents in one workspace
Each agent gets a dedicated computer
It is even possible to create a structure where one agent manages another agent

In other words, you are turning agents into a full operating ecosystem, not just standalone tools.

Tools that are effectively essential for operating agents

Composio removes the connection barrier

The biggest problem when operating agents is authentication and security.

You need to connect Gmail, Slack, Notion, GitHub, and many other tools, and this process is both annoying and risky.

Composio solves that.

Connect thousands of apps with one connector
Handle authentication
Handle tool calls
Reusable connectors

In short, it is a standardization tool for agent infrastructure.

Agent Mail gives agents a sense of “personhood”

If you give an agent its own email address, it feels more like a personal assistant to the customer.

For example, if the agent’s name is “Mia,” then Mia can send and receive emails directly.

This is not just decoration.

It is a mechanism that changes the customer experience.

Obsidian is your second brain

Agents ultimately need context in order to work well.

That is why Obsidian matters.

People information
Project information
Task history
Notes
Conversation records

If you store this information in markdown, agents can respond much more intelligently.

If you attach a recording layer like Limitless, conversation history builds automatically and becomes a real personal knowledge base.

In other words, Obsidian is the memory layer for a personal AGI.

Recommended models and stack summary

Model selection should be kept simple

As of now, the recommended flow is:

High-performance general-purpose work: GPT-5.5
Lightweight tasks: GLM 5.1, Kimi
Long coding tasks: Opus 4.7

The key is not to become emotionally attached to any specific model.

Models keep changing.

So structure, connection, and operations matter more than the model.

If you summarize the stack in one line, it looks like this

Codex: The simplest and most permissive execution environment
Hermes: Stable agent operations
Orgo: Workspace based on dedicated computers
Composio: App connection standard
Agent Mail: Agent email system
Obsidian: Context repository
GPT-5.5: Primary model

Why the way agents live in Orgo matters

Virtual computers are not just convenient; they are scalable

Local devices like Mac minis are inconvenient to operate.

As customers increase, incident response also becomes difficult.

By contrast, cloud-based virtual computers offer:

Isolation
Easy deletion
Easy recreation
Better security
Separation across multiple customers

In other words, it is a system that can scale to 100 customers.

Workspace-level separation is the key

Create a separate workspace for each customer.

Then each customer’s agents and data are separated, and management becomes easier.

This approach is not just a UI choice; it is an operational philosophy.

The age of agents setting up agents

Why Perplexity, Exa, Context7, and X MCP matter

Agent setup depends on the latest information.

That is why external knowledge MCPs matter.

Perplexity MCP: Search for the latest documents and information
Exa: Real-time web search
Context7: Reflect the latest GitHub documentation
X MCP: Capture practical examples and case studies from X

This combination is effective because agents can research and set themselves up using the latest information.

This is a step beyond ordinary automation.

Running multiple sub-agents is efficient

It is not just one main agent doing everything.

You launch several small sub-agents to research separately, then gather their results.

This is faster and more accurate.

Operational stability ultimately determines whether the business survives

A watchdog is essential

Gateways, connections, and automation flows can fail at any time.

That is why a watchdog is needed.

A watchdog is a monitoring system that automatically recovers when problems occur.

This is not a fancy feature; it is a mechanism that protects revenue.

An alert system is necessary for debugging

If an agent causes a problem, it should send an email alert by itself.

That way, the operator can intervene immediately.

The core point is simple.

You need to know whether it is working
You need to know immediately if it is not working
You need to be able to fix it right away

Without these three, it is not a business; it is an experiment.

The most important point rarely discussed elsewhere

The real essence of this business is not selling “AI technology”

Most people focus on agent technology, model performance, and automation demos.

But the real important thing is elsewhere.

It is selling an operational product that removes the complexity decision-makers deal with every day.

In other words, you are not selling AI itself. You are selling:

Context organization
Less follow-up
Clearer work visibility
Operational stability
Simplified customer communication
Faster decision-making

That is what you are selling.

The second key point is not to trust the amount customers say they need

Customers always say they want more features.

But in reality, one to three is often enough.

So in the beginning, you should offer something that appears broad, but design the actual operation to be narrow and solid.

Understanding this difference reduces token costs, development difficulty, and customer stress.

The third key point is that the competitive advantage of an agent business is security and stability

While everyone talks about what it can do, the part that makes money is how to operate it without breaking.

Customers are buying trust, not demos.

That is where the real competition happens.

Summary: how to think about this trend now

The AI agent market is moving from feature competition to operational competition

Going forward, the question is no longer simply whether you can build an agent.

What matters is:

Who you sell to
What problem you solve
How quickly you deploy it
How stably you run it
How easily you package it so customers understand it

From this perspective, a solo AI agent business is a very strong model.

In particular, providing AI employee-style services to legacy industries has strong potential for continued expansion.

< Summary >

The core of a solo AI agent business is not the AI technology itself, but selling “AI employees” and ensuring operational stability.

Legacy industries such as marketing, law firms, insurance, manufacturing, wholesale, and real estate are better suited than healthcare or finance.

Customers want business outcomes and efficiency, not tokens or models.

The success factors are content-based customer acquisition, rapid initial setup, an operational stack such as Obsidian, Composio, Agent Mail, and Orgo, and maintaining stability through watchdogs and alerts.

[Related posts…]

*Source: https://maily.so/josh/posts/knrj1pn1rld?from=email&mid=gz26k3x9wo3