HIRING
The questions that actually separate a real hire from an expensive toolkit — who is building it, who is on the hook after week one, what stays your call, and the twenty-minute check you can run before you sign.
The way this usually goes: the demo looked sharp, the person on the call was easy to like, and the pitch said the busywork would finally leave. Six weeks later you have a login, a Slack channel, and a workflow that still needs you every afternoon. When you asked who owns it if you stop paying, the answer was a help article.
Nothing illegal happened. Nothing was even hidden — it was in the order form. You just never asked the questions that would have surfaced it.
Choosing who builds and runs your AI workflows is not really a software decision. Almost everyone showing you a demo can make a message send. What separates them is structural: who does the work, who is on the hook after week one, and what you still decide. Those things are knowable up front. Almost nobody asks.
Not "bad AI." The model is the part everyone will claim is fine. The failures cluster elsewhere.
You became the operator. The most common and most expensive problem. They built it. You run it. Your evenings are now prompt edits and "why didn't this send." The hire you thought you made was a toolkit with a kickoff call.
Nobody is accountable for whether it works. It launches. Everyone congratulates each other. Nobody defined what "the job left your plate" meant. Twelve weeks later you have no idea if the quiet quotes moved, because nobody agreed what to count.
The person who impressed you is not the person building it. You met sales. Build goes to a contractor. "Success" goes to an account manager who relays messages between people who never speak. Every round trip loses the thing you said on the first call — how your shop actually runs.
After week one, the calendar goes quiet. Launch was the product. The weekly run was implied. Implied work is the first thing to disappear when their next demo is due.
You cannot leave. The conversations, the lists, the "what ran" live in a box that does not export. Your monthly fee is no longer buying a hire. It is buying continued access to your own work.
Ask these. The answers sort candidates faster than any demo.
You want names and roles, and you want to know how many handoffs sit between your "that doesn't sound like us" and the person who changes it.
Neither answer is automatically wrong. A team can be excellent. One person can be excellent. The structure tells you what you are paying for and how fast a correction happens. If the answer is "our team handles that," you are likely funding coordination rather than the workflows.
Every credible arrangement includes something ongoing. What you are testing is whether they can say it in a sentence you could repeat to a partner.
A good answer is a cadence: we run these two jobs, you get a note, here is what you do when something looks off, here is what you never have to open.
A bad answer is a dashboard tour, a "you'll have full visibility," or a support inbox. Visibility is not a run. A queue is not a hire.
If you want the product version of that cadence, it is on what a standard hire includes: two workflows, named; we run them; you get the weekly note.
The useful answer is a short list. Pricing a job. Promising a window. Anything that spends trust you cannot get back.
The useless answer is "you stay in control of everything," which means you will be in the tool forever, or "we handle all of it," which means they have not met a local service business.
You should like the fence. A hire without a fence will eventually send something you have to apologize for. A hire with a fence that includes every message will never take the job.
Ask it that bluntly, and get it in the agreement.
A good answer is specific: here is how you take the transcripts, the lists, the record of what ran. It might come with an honest caveat — you would need someone to stand the workflows up elsewhere — and that is fine. Honest specifics are the signal.
A bad answer reframes the question. "You'll never want to leave." "It's all in the platform." Those are answers about how leaving is unlikely, not about what happens if you do.
Not "we'll automate your front office." After-hours missed calls. Quiet quotes. Review asks. Named the way you say them, in writing, before build.
A shop that cannot name the jobs will discover them on your clock. We wrote about that failure mode — and the four offers that hide inside the word AI — in what you're actually buying when you hire an AI employee.
This is worth its own heading because it is the part owners either fear or skip.
You decide the jobs. You decide the voice, by approving the first stretch. You decide the fence — what never goes out without you. You decide whether a weird case is a yes, a rewrite, or a "I'll take this one."
You do not decide, if the hire is real, whether Tuesday's review asks go out. You do not live in the inbox of almost-sent messages. You do not become the intern for your own automation.
If that split is not clear in the first conversation, it will not become clear after you pay. Ask them to say it back in their words. If they cannot, they do not have an operating model. They have a demo.
Before any proposal:
None of this requires a technical background. All of it is more predictive than a screen recording.
Worth saying plainly, because articles like this are usually written by the people they recommend.
Hire a larger operation when you have multiple locations, a real ops team, and you want a dedicated account manager and a bench on call. Those are real needs. A small shop is the wrong shape for them.
Buy the widget or the DIY toolkit when you already have someone who likes systems, the job is narrow, and you wanted a tool, not a hire. That is a clean purchase if you are honest about who will run it. It is a bad purchase if you were trying to get the nights back.
Stay with a person when the work is judgment, presence, or a role you have not named. Whether an AI employee is worth it is the longer version of that fork.
A small shop that builds and runs the workflows makes sense when you want two named jobs gone without funding a layer of account management, you would rather talk to the person who will still be there in week six, and you value a correction this week instead of a ticket next month.
There is no universally correct answer — only a correct answer for the size of the mess and how you want your weeks to feel. Anyone who tells you their model is right for everybody is selling, not advising.
Demo quality is table stakes and nearly everyone clears it. What varies is who builds it, who runs it after launch, what you still decide, and whether you can leave. Ask for names, ask for week six, ask for the fence, ask for the two jobs in writing, and agree on what "off your plate" means before anyone connects a tool.
If you want a straight conversation about your shop — including an honest read on whether a person, a widget, or a bigger operation would serve you better — name the job. Twenty minutes. We will tell you if we can take it. And if you want the offer map before the call, start with what you're actually buying.
Founder of Revauri. Builds and runs AI employees for local service businesses — the jobs you'd otherwise hire a person to do.
