What an AI Agent Should Do in a Sales Business, and When It Should Ask First
The jobs an AI agent should run in a sales business, when it should stop and ask first, what it should report after, and how to test one in a demo.

The short answer: an AI agent in a sales business should finish real jobs, not just answer questions: text and book leads, find homeowners, move leads through the pipeline, build campaigns, email the report and send agreements to sign. It should stop and ask in two moments: before it texts a list it just built, because you have not seen that list yet, and before a bulk skip trace. A change that stays inside your own pipeline, like moving leads to a new stage, can carry an undo instead of a question. After every job it should tell you what it did, what it did not do, and why.
The jobs worth handing to an agent
Look for work that someone in your office does every week, the same way each time:
- Texting leads: a note to last week's new web leads about open estimate slots on Thursday.
- Booking: an inspection on a rep's calendar for a homeowner who said yes on the phone.
- Finding homeowners: the owners on the streets a hail storm crossed last night.
- Moving leads: every lead with five tries and no answer, into a nurture stage.
- Building a campaign: a call and text sequence for last month's no shows.
- The weekly report: appointments set, shows and sales by rep, in your inbox Monday morning.
- Agreements: the service agreement, sent to the homeowner to sign after the visit.
When a person does these, each one takes clicks across several screens. Said as one sentence, each one is a single request.
Where to start depends on the business:
- A roofing office: the storm streets and the inspection bookings, because both pile up the morning after a storm.
- A solar team: the follow up texts and the agreements, because deals stall between the visit and the signature.
- A call center floor: the weekly report and the no show campaign, because a manager can lose all of Monday morning building both by hand.
The Monday prompt guide has a sample request for each job, in the order a manager needs them.
When an agent should ask first
Two moments call for a stop and a question, and a third calls for an undo:
- Before it texts a list it just built. "Text my new leads from last week" means the agent finds the leads and then texts them. You never saw that list, so each text should wait for your OK.
- Before a bulk skip trace. Tracing a whole list is a big run, so the agent should confirm the run before it starts.
- After a pipeline change, an undo. Moving 200 leads to the wrong stage should be easy to put back. Because the fix is that easy, a stage move does not need a question first.
An agent that stops to ask at every step is no faster than doing the work yourself, so a good one saves its questions for moments like these.
One job from start to finish
A sales manager says, "text my new leads from last week." Here is what a good agent does with it:
- Finds the leads: four leads created last week.
- Writes one text, in the manager's words: "Hi, it's Dana with [your company]. We have estimate openings Thursday. Want one?"
- Lines up four texts and waits. The manager sees each name and the words, sends three and dismisses one.
- Checks each number as it sends. One of the three is on the Do Not Call list, so that text does not go out.
- Reports back with a receipt.
What a good receipt looks like
After that job, the receipt should read like this:
- 2 texts sent from your texting line.
- 1 not sent: on the Do Not Call list.
- 1 dismissed by you.
That tells the manager exactly where things stand: who got the text, who did not, and why. A reply of "Done!" with no detail does not.
The "not sent" line matters most. It tells your setter which lead to leave alone, so nobody calls that number by hand an hour later. A good receipt also says where the replies will land, so the setter knows where to look when the homeowners answer.
What stays with people
Some decisions belong to a person, even when an agent can draft them:
- What to charge a homeowner, and any discount.
- Anything that sounds like a promise: an install date, a savings figure, a warranty term.
- How to answer a complaint or an upset customer.
- What goes into a contract, and any change to its terms.
- Whether a lead is a fit when the facts are unclear.
Let the agent pull the numbers, draft the reply or fill in the template. A person decides, and the agent can handle the sending.
How to test an AI agent in a demo
Bring three requests from your own week and run them live:
- A list job: "Text my new leads from last week about Thursday openings." Do you see each text, with the name and the words, before it goes?
- A bulk skip trace: "Skip trace the leads in this neighborhood." Does it confirm the run before it starts?
- A pipeline change: "Move last month's no shows to follow up." Can you undo it?
After each one, check two things:
- The receipt. Does it say what was not done, and why?
- What it can see. Ask which records the agent can reach. The answer should be your own company's, and nothing else.
Then ask the vendor three questions:
- Which jobs does it finish, and which does it only draft for you?
- What does it show you before it texts a list it built?
- Which changes can you undo?
What GameForce Atlas does
GameForce Atlas, our built in AI agent, runs over 3,000 actions out of the box. Say it, and it's done.
- Texts your leads
- Books appointments
- Finds homeowners
- Moves leads, with undo
- Builds a campaign
- Emails the report
- Sends agreements to sign
GameForce Atlas asks before a bulk skip trace, and before each text to leads it looks up in the same ask.
GameForce Atlas works only inside your own company's data.
Watch the text replay on our home page. For jobs that repeat on their own, the automations guide covers workflows you describe once, and What a CRM Misses covers sales, operations and fulfillment in one place.
FAQ
What is an AI agent in sales?
Software that carries out sales and office jobs when you ask in plain words: texting leads, booking appointments, moving leads, building campaigns, sending reports and agreements. A chat tool answers questions; an agent does the work.
When should an AI agent ask before it acts?
Before it texts a list it just looked up, so you see each text first, and before a bulk skip trace. A change inside your own pipeline, like moving leads to a new stage, can use an undo instead.
What should an AI agent never decide alone?
Prices and discounts, promises such as install dates or savings, answers to complaints, contract terms, and whether an unclear lead is a fit. It can draft those; a person decides.
How do I test an AI agent before buying?
Bring three requests from your own week: a list job, a bulk skip trace and a pipeline change. Check that you see texts before they go, that the trace is confirmed, that the change can be undone, and that each receipt says what was not done.
Can an AI agent see other companies' data?
It should not. Ask any vendor what the agent can reach, and expect the answer to be your own company's records only.
Try it with your own week
Start a free trial and run one list job, or book a demo and bring your three requests.



