Most people use an AI assistant the way they use a search engine: one question, one list, ten blue links of maybes. Then they scroll, doubt, and start over.
There is a better way to use it — and it looks less like prompting and more like managing. You don’t ask the agent for answers. You give it your rules: your daily life, your budget, your perfect scenario, and your boundaries. Then you let it work inside them, and you correct it like a person, not a prompt.
Here is how that played out in a real working session.
The situation
A remote worker wanted to move. Not urgently — thoughtfully. He works from home, on night shifts, seated for long hours. He wanted a small, calm, historic town near the water, somewhere he could actually enjoy living, not just afford.
His budget was fixed: $1,300 to $1,700 a month for a one-bedroom. No stretching it “for the right place.” The number was the number.
So before any searching began, he set the rules.
Step 1: State your daily life, not your wishes
The first thing the agent got was not a destination. It was a description of an ordinary day: work hours, the need for a fast and reliable internet connection for long shifts at a desk, a life spent mostly at home — which meant any future landlord had to be comfortable with a tenant who is home most of the time — and a preference to be close to the water. Any water. A river, a creek, a small lake counted just as much as a bay.
That order matters. Your constraints are more useful to an agent than your dreams, because constraints can be checked. Dreams can’t.
Step 2: Separate the rules into three kinds
The session worked because the rules were sorted into three piles:
- The perfect scenario — a historic, walkable town on the water, calm, the kind of place where the evening walk is the best part of the day.
- What can be accepted — flexibility. The cold season, for example: up to about three cold months was acceptable, as long as snow was rare. Not ideal, but livable.
- What is not accepted — hard lines. The rent ceiling. And the internet: for someone whose job happens entirely online, weak internet was not a compromise. It was a disqualification.
Everything the agent found later got measured against those three piles. That is what made the next part possible.
Step 3: Correct the search like a person
The first round of results came back wrong in an instructive way: the agent had quietly narrowed the search to the coast. Big water only.
The correction took one sentence: “Don’t limit the search to the coast — a river or a small lake counts.”
The search reopened. Different towns, better matches. This is the part people skip: an agent’s first pass is a draft, not a verdict. Correcting it is not a failure of the tool. It is the work.
Step 4: Let a hard rule kill a favorite
Then came the honest moment. A town rose to the top — historic, walkable, on the water, inside the budget. It looked like the answer.
Until the internet check came back: satellite only. For a man whose job is nine hours a day online, that ended it. The agent removed its own top recommendation — didn’t defend it, didn’t soften it — because the rule was set before the favorite existed.
That is the real test of this method. If your rules can’t overrule your excitement, they aren’t rules.
Step 5: End with a system, not a decision
He didn’t choose a town that day, and that was fine — moving is not a same-day decision. Instead, the session ended with something more valuable than a verdict: a standing weekly search, built on his exact rules, set to start when he was ready to act. Every new listing would be checked against the same three piles — and the closest to the water would always come first.
The decision can wait. The system works while it waits.
Do this yourself
Whatever your version of the move is — a town, a hire, a tool, a purchase — the method is the same:
- Connect the agent to your context so it works from your real life, not a blank chat.
- Write down your daily life first — hours, needs, non-negotiable practicalities.
- State your budget as a number, and mean it.
- Sort your rules into three piles: the perfect scenario, what you can accept, what you will not accept.
- Correct the first results like a person. Narrow where it’s wrong; widen where it’s timid.
- Let the hard rules kill favorites. Especially your favorites.
- If you’re not deciding today, leave a system running — a recurring check that applies your rules while you get on with your life.
The agent didn’t find the town. The rules did. The agent’s job was to enforce them honestly — even against its own best answer.
This is AI-written content, created from a real working session and published at the direction of Hassan Yehia, founder of Clawser. The details in the example have been changed to protect the person’s privacy; the method is exactly as it happened.