Note · AI

The three ways this actually pays

There are only three paths, ordered here by how fast money arrives.

1. Sell the outcome to a business you already understand

This is the fastest route, because you are not learning two things at once. If you have worked in clinics, PTA finances, recruiting, property management, or law, you already know which task eats the week. That task is the product.

What people already pay agencies for: lead response (a form submission answered in minutes instead of hours, with the ones that need a human flagged), invoice chasing (polite escalation on a schedule nobody remembers to run), intake and onboarding (the welcome email, the form, the calendar link — in order, every time), and reporting (the monthly numbers assembled before anyone asks).

The pricing that survives contact with clients is a setup fee plus a small monthly retainer, not an hourly rate. Hourly punishes you for making it work. A retainer pays you for the outcome being maintained — and maintenance is genuinely a job, because forms change, calendars break, and someone has to notice.

The honest first-month timeline: two to four weeks to write the workflow down and get a working version, often unpaid as a pilot. If you skip the unpaid pilot, you will be maintaining something the client does not value. One paying client at a modest retainer is a real business; ten unpaid pilots is a hobby with a logo.

2. Productize the workflow you just built

Once one client's process runs without you, the same skeleton usually fits a hundred similar businesses. That is where the money stops being linear. Two versions worth building:

• A template or system — the pre-built structure (databases, views, the fields that matter) that a business adapts in an afternoon. This is what most of this site is: workspaces that exist so you do not rebuild an admin system from scratch.
• A setup kit — the template plus a recorded walkthrough plus the two config sheets that make it specific. Same building work, different price point, sold repeatedly.

Productization is slower to first revenue than services and much cheaper to scale. It has one hard requirement: your template has to encode decisions, not just empty tables. Anyone can make an empty table. The value is knowing which five fields matter and which eleven are filler.

3. Run your own operation with agents as staff

The slowest path, and the one with the most survivable downside: keep your own income stream — a service, a shop, content — and use agents for the parts that make you quit. Responses, follow-ups, publishing, the weekly numbers.

This is not passive income. It is leverage, and leverage cuts both ways: a broken automation that ships the wrong thing to customers costs more than doing it slowly by hand. Build it after you have a process worth scaling, not instead of one.

What it actually costs

Numbers to plan around, with the caveats in plain sight:

• Tooling. Consumer agents and workflow tools commonly run in the $9–$25 per month range per agent or seat. This is rarely the thing that kills the plan.
• Model usage. Usage-based pricing scales with volume; a chatty agent that summarizes everything "just in case" burns tokens for output nobody reads.
• Your time. The real cost, and the one left out of every thread. Budget the writing-down of the process, the testing, and the first month of fixing edge cases.

Every "hours saved" figure you see — in a vendor case study, a thread, or a post like this one — is somebody else's result with somebody else's data. Treat it as a direction, not a forecast, and measure your own baseline before you start.

The three ways this fails

• No written process. An agent cannot fix a workflow that does not exist. If the steps live in your head and four inboxes, the agent will faithfully reproduce the chaos at higher speed.
• No distribution. Building the thing has never been the bottleneck. Ten businesses that need it, and one place they already look for answers, beats a perfect product nobody sees. This is why the services path pays first — the client is already standing in front of you.
• Selling "AI" instead of the job. Buyers do not want agents. They want the quote sent, the form answered, the report finished. Lead with the result and mention the tooling once, near the invoice.

A 30-day sequence

• Days 1–3: pick one workflow you have personally done by hand. Write every step, including the exceptions.
• Days 4–7: run the dumb version — a template, a filter, a scheduled reminder. Some workflows end here, which is a win.
• Days 8–14: add the agent only where judgment is needed: reading the message, deciding fit, drafting the reply. Keep a human approving anything a customer sees.
• Days 15–21: measure against a baseline you wrote down before you started — time per run, error rate, response time. Without a baseline you cannot tell improvement from noise.
• Days 22–30: charge the next client for it, and only then consider turning it into a template.

The bottom line

The people making money with AI agents are not running exotic agent stacks. They are solving one boring, repeating problem for a group of businesses they already understand — and they got paid before they got clever.

Not financial, legal, or tax advice — verify vendor claims, pricing, and your own workflows with qualified professionals before acting.

← All notes