Salary: $0/Month. Benefits: None. Meet the AI 'Employee' Running a Michigan Bookkeeping Firm's Front Desk

Dana Whitfield runs a bookkeeping practice in Grand Rapids, Michigan, with exactly one other human on payroll. When her front-desk person gave notice in March, she did what most small business owners would do: she drafted a job posting. Then she did something most wouldn't. She deleted it and spent a weekend building an AI agent instead.

I checked in with her every couple of weeks over one full quarter to track what actually happened — the dollars, the disasters, and the parts nobody puts in the product demo.

The Setup: Less Hiring, More Configuring

Dana's stack ended up being simpler than she expected. A scheduling tool with API access, an email automation layer, an invoicing platform, and a glue layer connecting them — plus roughly 20 hours of her own time teaching the agent her tone of voice, her client list, and the difference between a real emergency and a mildly annoyed customer.

Total monthly cost: about $180 across subscriptions. The part-time hire she'd been considering? Around $2,200 a month with payroll taxes. On paper, that's a 92% discount.

On paper is doing a lot of work in that sentence.

Month One: The Honeymoon

The first few weeks went almost suspiciously well. Appointment scheduling — the agent's strongest suit — ran flawlessly. Clients booked slots at 11 p.m. on Sundays. No-shows dropped because reminder emails went out like clockwork. Dana said the booking flow felt "like the office had a night shift for the first time ever."

Invoicing was nearly as smooth. Recurring invoices went out on schedule, and the agent flagged two late payments Dana would have missed for another week.

Then came the emails.

The Mistake That Almost Ended the Experiment

A longtime client wrote in with a frustrated, vaguely worded message about a discrepancy in her quarterly filings. The agent responded promptly — with a confident, polite, and completely wrong explanation of what had happened. It invented a reason. It apologized for something that didn't occur. And it promised a correction that was never coming.

Dana caught it an hour later because the client called her directly, confused. "That one email cost me more in goodwill than the agent saved all month," she told me.

The fix was unglamorous: she rewrote the agent's instructions to escalate anything involving client money, filings, or emotional language straight to her inbox. In AI terms, she added a human-in-the-loop trigger. In small business terms, she realized the agent couldn't be trusted near the furnace.

The Quarter-One Ledger

Here's the honest accounting after 13 weeks:

• Saved: roughly $5,400 versus a part-time hire, plus untold hours of scheduling back-and-forth
• Spent: about $540 in subscriptions and 30+ hours of setup, tuning, and cleanup
• Recovered: one client relationship, narrowly, after the email incident
• Gained: genuine 24/7 availability, which Dana says clients notice more than she expected

Net positive? Clearly. A drop-in replacement for a person? Not even close.

What the Agent Can't Do (Yet)

The pattern across the quarter was consistent: anything with a script, the agent crushed. Anything requiring judgment, it fumbled or overstepped. It never once forgot to send a reminder. It also never once noticed that a client's tone had shifted from annoyed to angry, or that a question about "the invoice" probably meant the one from October, not the one from last week.

Dana's verdict, three months in: "It's not an employee. It's a really good assistant who lies confidently and never gets tired. You have to supervise accordingly."

So Is This a Staffing Solution or a Liability?

Both, honestly — and the difference is configuration. Dana's setup works because she treats the agent like a new hire on permanent probation: high-trust tasks automated, low-trust tasks escalated, everything audited weekly. The businesses that get burned by "AI employees" tend to be the ones that hire them and walk away.

For US small businesses staring down labor costs, the math is tempting and often real. But the quarter-one lesson from Grand Rapids is simple: you're not saving a salary. You're trading it for a job you can't fire — the job of supervising something that works 24/7 and never admits uncertainty.

Dana says she'd do it again. She'd also hire a person the moment her business could afford one. The agent, she says, isn't the team. It's the scaffolding while the team is two people and a very tired owner.