The smallest AI stack that actually works for a producing agent

I ask a version of the same question in every workshop: how many AI subscriptions are you paying for? The median is four. The median number being used in a normal week is one, and half the time that one is being used as a slightly better search engine.

This is not a discipline problem. It is a design problem. Tools get bought because of a demo rather than a job, and a tool without a job has nowhere to live in a week.

Four jobs, in priority order

Start from the work, not the software. For a producing agent, four jobs absorb nearly all the recoverable time.

1. Capturing what was said. Calls, showings, listing appointments. Everything you learn and then partially forget between the car and the desk.
2. Turning that into follow-up. The note, the summary, the next three things, sent while it still matters.
3. Producing the written assets. Descriptions, updates, email, the market note you keep meaning to send.
4. Answering questions that would otherwise cost you thirty minutes of research.

One tool per job. A general assistant handles three and four. A transcription tool that plugs into your CRM handles one and most of two. That is genuinely it for the first ninety days.

A tool without a recurring job attached to a specific hour of your week is a subscription, not a system.

Attach each tool to a moment

Adoption fails in the gap between intention and trigger. The fix is to name the moment out loud, in the calendar, next to something that already happens.

• Recording starts when the car starts. Not when you remember.
• The summary and the follow-up go out before you leave the driveway at the next stop.
• Written assets are produced in one block, same time every week. Mine is a ninety-minute Tuesday block and it has not moved in three years.
• Research happens the moment the question appears, not on a list called "look into."

The order matters more than the choices

Agents almost always start with content generation because it is the most visible and the most fun. It is also the lowest-leverage of the four, because the constraint on your business has never been the volume of words you can produce.

Start with capture. It is unglamorous and it is where the compounding is. An agent who never loses what was said in a listing appointment converts better than one who publishes twice as much, every time, and it is not close.

What to add later, if ever

Once those four are habitual for a full quarter, the next additions worth considering are a scheduling layer that removes the back-and-forth, and something that watches your database for signals rather than waiting for you to remember. Both are real leverage. Both are also useless on top of a foundation that is not running.

The best stack I have seen this year belonged to an agent doing thirty-eight transactions with two tools and a habit. The worst belonged to someone with eleven subscriptions who could not tell me what happened after a listing appointment.