Write Down Every Screen Task Before You Hand It to an AI

October 01, 2026•7 min read

The most useful thing I told the room this week had nothing to do with which AI tool to buy. I told everyone to start writing down every task they do in front of a screen. Every task that involves a phone. Every repetitive click. The tools are changing fast enough that the list will outlast any tool you pick today.

Start With the List, Not the Tool

Most people approach automation backwards. They pick a tool, then hunt for something to point it at. The order that works is the reverse. For a week, log the tedious work: the reports you pull, the emails you send that look the same each time, the notices, the reminders, the follow-ups.

Once a task is written down step by step, it is a procedure. A procedure can be handed to a person, to the software you already own, or to an AI agent. Much of the routine work in property management can already be automated if you think the process through: automatic texts and emails to residents, notices, even printing and mailing. The bottleneck is rarely the software. It is that the process only exists in someone's head.

This is the same move behind REAP Principle 44 on systemizing your business. Write it down and it stops depending on you.

A Source of Truth Beats a Smarter Model

Every AI tool I use gets better when it has one place to read from. That place can be a shared drive, a document, or your property management platform. What matters is that the rules for how your company works, and the facts about your buildings, live somewhere the agent can read every time.

If your platform has an open connection, the agent can pull work orders and reports directly. If it does not, there is a workaround: have an agent log in through a browser to the owner portal, pull the reports each day, and drop them in a shared folder. Slower, but it gives the agent a clean source. The larger lesson is in why AI multiplies whatever you plug it into. A tool connected to good data looks brilliant. The same tool connected to nothing looks average.

One Agent, One Job, One Shared Log

The question that came up most was how to manage several AI tools without them stepping on each other. My answer is the same one I would give a human team. Give each agent a role and do not let anyone work outside it.

The setup is simple. Every agent reads a shared task file before it starts and writes to it when it finishes. The file grows longer over time, so each agent can see what has already been done. The rules are short: do not repeat each other's work, and do not take a task that carries someone else's name. Email replies belong to one agent. Report pulling belongs to one agent, and everyone else can read the output. If two agents share a job, you get duplicates and conflicts.

I use the same logic with the people in the business. Tasks have names on them. Anything without a staff member's name is an agent's task. That is the whole system.

Test the Tools on the Same Job, Then Pick One

If you are unsure which tool is right for a task, run a small contest. I gave two AI tools the same photo of a damaged unit and asked each to write up everything wrong with it. Then I handed both reports to a third tool and asked it to judge. One had filled its report with assumptions. The other described what it saw, and finished about three times faster. I picked the second one and it now runs that job on a schedule, a few times a day. The other tool does not touch that job anymore.

I ran a bigger version the other day. Several tools took the same research prompt. Each was told to save its work in its own named file, and none was allowed to read the others. At the end one tool combed through every folder and reported where they agreed. That told me which platform to assign to which type of task going forward. The goal is not to run a zoo. It is to choose the right platform for each lane, then let it stay in that lane.

Why Vacancy and Leaks Are the Reason to Care

I showed the room a dashboard I built to see every building at once. Units are color coded: red is vacant, with the number of days each has sat empty. That number is the first thing I look at, because a portfolio is under constant attack from vacancy.

A vacant unit is not just lost rent. It is also a unit with nobody paying attention. I was at a vacant building on Sunday where a ceiling was leaking. I put a one gallon bucket under it and it filled within a minute. I emptied it six times. A gallon of water weighs a little over eight pounds, so that was roughly fifty pounds of water in a few minutes, pressing on drywall that was already soaked. Left alone, that weight builds until the ceiling comes down, and the water moves floor by floor until it reaches an occupied unit and a resident calls.

The point is that a good resident is also your early warning system. Some residents will even fix small leaks themselves. Others will damage a unit. As the landlord, you cannot control the market, but you decide who you place from the pool that is available. The quality of the landlord shapes the quality of the residents, and the quality of the residents shapes the quality of the building.

That is why I pointed an AI agent at leak detection with one job: watch for the problem, read the work orders, and flag what needs attention. It is a narrow role, and that is why it works.

The Takeaway

Do not try to beat the machine. There is an old American folktale about a man who insists he can out-work a machine, and it does not end well for him. Hire the machine instead. A paid AI plan costs a small fraction of the person hours it replaces, and the work it does can add up to several people's worth.

Three moves for this week. First, write down every task you do in front of a screen or on the phone, and note which ones repeat. Second, choose one task and give it to one tool with one written rule set, then watch it for a week. Third, put the results somewhere a second person can see them. If you want to see what an empty door does to your own numbers, run your own return on equity on the calculator and put vacancy days into the picture.

If you find you have nothing to hand an AI, that is not a win. It means you are not doing enough yet. Pick up something bigger, or spend the time somewhere that matters more.