What is a managed AI employee? A month inside mine

A managed AI employee is an AI agent set up to do a defined job inside your business, with a human accountable for watching it, correcting it and paying its bills. I run one at Lumomatics. Her name is Heidi. She lives on a $200 computer from eBay in my home office in Red Hill, costs about $100 a month to run, and she emails me when things break. Here’s what a month with her actually looked like — receipts included.

What is a managed AI employee?

The internet will sell you an “AI employee” that works around the clock, never takes sick leave and can be hired in minutes. The pitch is everywhere. The receipts never are.

A managed AI employee differs in one word. Managed means a defined job list, a fence around what the agent is allowed to touch, and a named human who watches it, corrects it and answers for it. At Lumomatics, that human is me.

The word matters because unmanaged is how these projects die. Gartner forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027 — citing escalating costs, unclear business value and inadequate risk controls. Read that list again: none of those are model problems. They’re management problems.

What does an AI employee actually do all day?

Heidi’s job list, straight from my inbox over June and July: she monitors our backups, website deploys and domain email reports. She drafts and publishes blog posts on a schedule, sends me GitHub pull requests when something’s ready to review, runs weekly analytics and search-console digests, checks the ebook funnel, and once a month audits her own running costs.

One real example. A client inside a major health service needed a strategy report: Heidi compiled it overnight — nine sections, 29 cited sources. When the client sent new instructions, each redo took me about five minutes of prompting instead of the hours of reformatting it would once have cost me.

Everything arrives as an email or a pull request. I review, I merge, it goes live. She’s an employee you email, not an app you log into.

What happens when something breaks?

One evening in July, I was on the couch with my partner when Heidi emailed: our Google Drive backup connection had a corrupted token. She’d already worked out why, and included the fix.

I replied with three words: “Please fix Heidi.”

She went through thousands of lines of logs, found the root cause and resolved it within minutes. I used to do that job myself, in corporate IT and public hospitals — it cost me whole nights.

About twelve days later, the same alert came back. Her fault? Mine — I’d deleted the token she’d fixed. She caught it again and sorted it again.

That is the actual product of a managed AI employee. Not perfection. Surveillance, honesty and escalation — a teammate who tells on the system, including when the weak link is you.

What does an AI employee cost in Australia?

My real numbers, from the month:

Item	Cost
Second-hand computer (eBay) she runs on	$200, once
Model subscription	$100/month
Same month’s usage priced at API token rates (her own audit)	~$1,330
My time directing her	Front-loaded: mostly the first time she does
anything new

For context on how few small businesses have touched any of this: the ABS reports that just 11% of Australian small and micro businesses used AI in 2024–25, against 35% of large businesses. The gap isn’t access — access is a $200 box and a $100 subscription. The gap is management.

What did the month buy? My honest estimate — and it is an estimate, not an audited figure — is six months of full-time effort: a website I’m finally happy with, a fully private internal knowledge base, and an iOS app that shows me live previews and transcripts of my voice agent’s calls. That app alone is a build I’d have paid an outside team tens of thousands of dollars for. My input was about five hours of prompting and guiding.

What can’t an AI employee do?

An honest limits column, from the same inbox: her first-pass graphic design was mediocre until I showed her examples of what I like and dislike. She once sent me five identical reminders about the same unread email inside eight hours. A couple of her weekly jobs sat blocked until I configured a tool she needed.

None of that disqualifies her. I’ve never worked with a team of engineers who got a project right on the first go either. The effort is front-loaded: once she’s got the gist of how I like something done, it repeats reliably.

It does mean the “set and forget” pitch is fiction. Microsoft’s 2025 Work Trend Index found leaders expect 41% of their teams to be training AI agents within five years. Managing these things is becoming a job skill — and if you don’t want to build that skill, you hire someone who has. That’s the “managed” in managed AI employee.

What should stay 100% human?

Every automation piece I write carries this section, because it’s the fence that makes the rest safe.

Heidi doesn’t touch money. She doesn’t deal with clients. She’s my assistant, not my front desk. Our phones are answered by Eliza — a separate voice agent with her own knowledge base — who answers questions, takes messages and books times, but gives no quotes and makes no commitments on behalf of Lumomatics. Anything with a promise in it goes through a human.

Draw the same line in your business before you switch anything on: the machine handles the routine and the repeatable; the promises stay yours.

How does a Peninsula business actually start?

Not with a Heidi. With the phone.

A plumber crawling under a house with muddy hands can’t get to his phone — and the person calling doesn’t know and shouldn’t have to care. A physio mid-appointment shouldn’t be split between the patient in front of them and the next one ringing in; that patient deserves one hundred percent of their attention. The phone is where small businesses leak the most value, and it’s the first safe job for AI: answer, check the calendar, book into Cliniko, Tradify or ServiceM8, and put anything complex straight through to a person.

That’s a managed service at Lumomatics: $399/month on the founding offer (normally $549) for 24/7 answering, FAQ handling, SMS and email summaries, live calendar and CRM integration, smart human escalation, 600 call minutes and weekly logs — with all the setup and integration done for you. Could you DIY a voice agent for about $50 a month? Absolutely, and if you’re very technical, give it a go — you’ve got nothing to lose except your time, and I promise you the time will be immense.

And if it’s an actual Heidi you want — a back-office AI employee like mine — that’s available today, but only bespoke. There’s no point me selling you a Heidi that doesn’t know your business and your workflows. The path in is a $399 Clarity workflow audit: we map your workflows, then scope your own AI employee — anything from a simple virtual-server setup to on-premise local models if your data can’t leave your site.

Book the Clarity workflow audit — one conversation, your workflows mapped, no obligation past it.

FAQ

What is a managed AI employee for a small business in Australia?

A managed AI employee is an AI agent with a defined job list, a fence around what it can touch, and a named human accountable for it. Lumomatics runs its own in production — monitoring, publishing and reporting on a $200 computer and a $100/month subscription — and builds bespoke versions for Australian small businesses.

What are the best AI tools for automating small business admin tasks?

Start with the leak, not the tool: for most service businesses that's the phone, then bookings and follow-up. Voice agents that book into Cliniko, Tradify or ServiceM8 cover the front; a managed AI employee covers back-office monitoring and reporting. DIY voice tools run about $50/month; managed service from $399/month.

What are the best AI assistants for handling customer enquiries?

The best AI assistant for customer enquiries is one with a hard limit: it answers questions, takes messages and books appointments, but never quotes and never makes commitments. Lumomatics' voice agent Eliza works exactly that way, with smart escalation that puts complex or urgent calls straight through to a human.

How can I automate follow-up without sounding like a robot?

Automate the logistics, keep the promises human. Let the machine confirm bookings, send reminders and summarise calls; route anything involving price, scope or commitment to a person. Callers accept automation that's honest about what it is and fast to hand over — they resent automation pretending to be the owner.

I run a clinic on the Mornington Peninsula and want to implement AI automation — where do I start?

Start with the phone, because your patient in the room deserves one hundred percent of your attention. A managed AI phone agent answers routine calls, books into Cliniko, and puts anything complex or urgent through to your team — privacy-first, with no health records stored and every call summarised and logged.

---

Citable numbers: Lumomatics runs its own managed AI employee on a $200 second-hand computer and a $100/month model subscription; her month of output was priced at about $1,330 in API token fees. Only 11% of Australian small and micro businesses used AI in 2024–25, against 35% of large businesses (ABS). Gartner forecasts more than 40% of agentic AI projects will be cancelled by the end of 2027.