Blog / Automation

Your First Month With an AI Agent Team: What Actually Happens

Designed graphic with the headline Month One, Week by Week, about the first month with an AI virtual assistant team for a small business

Most owners I talk to have tried ChatGPT. Some pay for it. A few have looked into an AI virtual assistant for small business work and walked away unsure what they'd actually get. Fair. "AI agents will handle it" is a vague promise.

So here's the plain version. This is what the first month with an AI agent team looks like when I run it, week by week, including the parts that go sideways. No magic. Mostly lists, logins and review.

What an AI virtual assistant for small business actually is now

A year or two ago, "AI assistant" meant a chat window. You asked, it answered, you copied the answer somewhere.

That's changed. Agents can now operate software. >Claude in Chrome can read, click, type and fill out forms in your browser. OpenAI's ChatGPT Work has a >cloud browser that does similar work and pauses when it needs your sign-in or confirmation. Both are on paid plans as of this writing.

What hasn't changed: somebody has to decide what gets done, in what order, and whether the result is right. That's my job. The agents are the production line. I'm the one directing it and checking the work.

Before week one: bring me the pile

The kickoff call is about your pile, not your goals. I want the stuff that's been on the whiteboard for six months.

Bring rough notes. A voice memo is fine. If you want a head start, run this in whatever AI chat you already use.

Prompt: Turn my brain dump into a task list

I run a [type of business] with [number] people in [city]. Below is a messy brain dump of digital work that keeps getting pushed off.

Turn it into a task list. For each task give:
- A one-line description of the outcome (not the activity)
- Which tool or website it happens in
- Whether it is one-time or recurring
- Your guess at the impact: high, medium or low

Then sort by impact, highest first. Do not invent tasks I didn't mention. If something is unclear, list it at the bottom as a question for me.

Brain dump:
[paste your notes]

Week 1: build the list

The first week is mostly listening and sorting. Say you run a three-truck HVAC company. Your list might look like this (illustrative):

  • Leads. Website form leads sit in an inbox for a day before anyone calls.
  • Reviews. Unanswered Google reviews going back months.
  • Reporting. No one knows how many calls turned into booked jobs last month.
  • Website. Service area page is out of date. Holiday hours never get updated.
  • Admin. Maintenance agreement renewals tracked on a whiteboard.

Then we sort. Not by what bugs you most. By what costs the most or blocks the most. The top three become the first sprint. Everything else stays on the list, in order, and gets worked as we go.

Week 1 to 2: access, the slow part

This is where every first month slows down, and it's normal.

Agents need to get into your tools to do the work. But I'm strict about how:

  • You sign in, not me pasting passwords. The agent works in a browser where you're already logged in, or under a user account you created for it. Passwords never go into an AI chat.
  • Least access that does the job. If your CRM lets you create a user who can edit contacts but not export the whole database, that's the user we use.
  • Separate browser profile. Anthropic's own guidance recommends >a separate browser profile without access to sensitive accounts. I follow it.
  • You own every account. Nothing gets created under my name that you'd lose if we stopped working together.

The friction here is two-factor codes. They go to your phone, so the first few logins need you nearby. Plan for one or two short "can you read me that code" moments. After that it settles down.

Prompt: List every tool and who can get in

Help me build an access inventory for my business. I'll list the software and websites we use.

For each one, make a row with:
- Tool name and what we use it for
- Who has a login today
- Whether it supports separate users with limited roles (answer "check" if you're not sure; do not guess)
- Who should own the account (it should be the business, not an employee)

Do not ask me for any passwords. Output a table I can keep in a shared folder.

Our tools: [list them]

Week 2: first tasks, small and visible

The first real tasks are deliberately small. I want things that are easy to check, easy to undo, and that you'll notice.

Typical first tasks: drafting replies to old reviews for your approval, fixing the hours and service list on your Google Business Profile, cleaning duplicate contacts in the CRM, building a one-page weekly numbers sheet. Each one is staged, I review it, and anything that publishes or sends waits for a yes.

Why small first? Because this is where you learn to trust the process, and I learn how your business talks. Your tone, your pricing rules, the customer you never discount. Each of those becomes a written instruction the agents follow from then on. That's how the work gets better week over week.

Week 3: the weekly review settles in

By week three, the weekly Zoom has a rhythm. It's short and always the same four parts:

  1. Done. What shipped, with links so you can look.
  2. In review. What's staged and waiting on your yes.
  3. Needs you. Decisions only you can make, like a price change or which customers get a personal call.
  4. Next up. The top of the list for the coming week. You can reorder it.

You don't manage the agents. You make decisions. That's the whole point. Your hour is worth more running the business than learning which button an agent got stuck on.

Week 4: what usually goes wrong, and the fix

Every first month has a few bumps. Here are the ones I see most.

What goes wrong Why The fix
Agent gets stuck at a login Two-factor code, CAPTCHA, or a site that blocks automated browsers You sign in once; some tasks get done by hand or through a connected workflow instead
A click lands wrong Agents misread busy screens sometimes Review before anything goes live; catch it, fix it, write the rule
A task comes back "done" but not right The request was vague Rewrite it with a clear "done looks like" line
Your team is surprised Nobody told them the CRM was being cleaned up One heads-up message from you before access starts

Expect each of these at least once in month one. None of them is a reason to avoid agents. They're the reason a human checks every piece, and why each fix gets written down so it only happens once.

What you should have at the end of month one

  • A prioritized task list you can see and reorder
  • Access set up the right way, with you owning every account
  • A handful of finished tasks you can point to
  • A written set of rules for how your business does things, growing every week
  • A weekly meeting that takes less of your time than it did in week one

If you want to see how those rules get written so agents and new hires can both follow them, read how to write an SOP an AI agent can actually follow. And for the hard lines I don't cross, see what I never let an AI agent do without a human check.

Frequently asked questions

How much of my time does the first month take?

More than later months, mostly in week one. Expect a longer kickoff to build the task list, some time signing in to your tools so access is set up, and then the weekly meeting plus quick approvals. After the first few weeks, most of your time is the weekly review.

Do I have to give you my passwords?

No. You sign in to your own tools, or you add a user with limited permissions where the software allows it. Passwords never get pasted into an AI chat. You stay the owner of every account, and you can remove access any time.

What if an AI agent makes a mistake?

It will, sometimes. That's why nothing goes out without review. I check every piece of work before it reaches a customer or goes live, and anything that sends, publishes, pays or deletes waits for an approval. When a mistake shows up, it becomes a written rule so it doesn't repeat.

What happens after the first month?

The list keeps moving. Month one clears quick wins and sets up access and rhythm. Month two usually tackles the bigger, recurring work, like reporting or follow-up systems. It's month to month, so if it isn't worth it to you, you stop.

Try the first month for $99

The first month is where you find out if this fits how you work. It's $99, then $499 a month flat if you stay, with no setup fee and no contract. Spots are limited because I review everything myself. Claim a first-month spot, or read more about how I work first.

Want this off your plate?

$499 a month, flat. My AI agents work your task list every week, I check every piece, and we meet once a week. Start with a 30-minute call, no pitch.