What an AI crew does all week, and what it gets wrong

"AI agents" sounds like a magic box. It isn't. A crew is a small, managed team with written rules, a supervisor, and people who approve anything that matters. Here's what one actually does in a week, including the parts that go wrong.

The team

A crew has a few specialists and at least one coordinator (we call it the foreman); bigger setups have more than one.

• The foreman reads every request first, hands it to the right specialist, and checks the result against your playbook. When something is risky, the foreman asks a person.
• The front desk answers calls and texts, day and night, and sizes up the job.
• The scheduler books the right tech by skill, zone and the customer's best time, inside the field-service app you already use.
• Follow-up chases estimates on your cadence and asks happy customers for reviews.

Bigger shops add a bookkeeper for invoices and collections, or a memberships agent for renewals.

Every agent works from the same playbook: your price book, service area, on-call rules and tone of voice, written down in the first week.

Monday to Sunday

Monday, 6:41 a.m. A tech calls in sick with six jobs on his board. The scheduler moves four jobs to two techs with open blocks and one to Thursday at the customer's preferred time. One job is a water-heater install that needs a licensed plumber, so the foreman asks the office before moving it. By 6:52 the day is rebuilt and three customers have new windows.

Tuesday, 9:00 a.m. A $14,800 roof estimate has been quiet for six days. Follow-up checks the playbook (follow up on day 3, 7 and 14, mention financing, never discount without the owner) and sends a short text in the owner's voice.

Wednesday, 2:14 a.m. An AC is out in Tampa, it's 88°F inside, and the caller's mother is 84. The playbook says that's an emergency, and after-hours dispatch costs $189, so the foreman asks the on-call tech. He says yes. The customer has a maintenance plan, so the fee is waived and a tech arrives at 3:05.

Saturday, storm day. A hundred and forty calls in six hours. The foreman ranks active leaks first, then no power, then estimates, fills the next 48 hours of capacity, and asks the owner what to tell the four leaks that can't fit before Friday. Every caller gets an honest status text. Nobody gets an arrival time the crew can't keep.

Where people stay in charge

The crew does the routine work. People decide anything that costs money or changes a promise:

• Emergency dispatch and after-hours fees go to your on-call tech.
• Discounts, refunds and prices outside your price book go to the owner.
• Anything the playbook doesn't cover goes to one of our engineers first, then to you.

You approve from a portal, a text or an email. Every decision is written down.

What it gets wrong, and how we catch it

Agents are probabilistic. They will make mistakes. So we design for the common ones:

• A misheard address. Fix: the front desk reads the address back and confirms it before booking.
• An over-promised arrival. Fix: the scheduler checks real capacity before giving any time window.
• A wrong price. Fix: agents can only quote from your price book. Anything else becomes an approval.

Every day, an evaluation agent of ours reviews every conversation against your playbook and flags anything that looks off. An engineer aims to fix flagged issues within one business day and logs every change. Your monthly report lists the mistakes found and the fixes made, alongside calls answered, jobs booked and hours saved.

What a week costs to run

Most of the cost is the voice. At about $0.12 a minute, a week with 250 calls averaging four minutes uses about 1,000 minutes, around $120 of voice. Texts are about a cent each. Our Crew package includes 1,000 voice minutes and 1,500 texts a month, and the Lab doubles that.

If you want to see what your own week might look like, answer five questions. You'll get a plan with your likely crew and the hours your office would get back.