What to Look For Before Hiring Your First AI Employee
How-to — — by Mahmoud Zalt
A pre-purchase checklist for your first AI employee: the job, the examples, the guardrails, the access, and the number you will judge it by.
Nobody hires a person by signing a contract and then wondering what to give them on Monday. Yet that is exactly how most first AI Employee hires go.
Sign up, poke around, ask it something vague, get something vague back, and quietly stop opening the tab. The tool takes the blame for an onboarding that never happened.
This checklist is the preparation, and it takes about forty minutes. **Sistava** is the platform I build, but every item below applies whichever one you pick, and doing them first is worth more than picking perfectly.
At a Glance
- 5
- Things to prepare first
- 40 min
- Total preparation time
- 3
- Real examples to gather
- 1
- Job, not a wishlist
One: pick a job, not a department
Your first AI Employee should own one repeated task with a clear right answer, not an entire function. "Handle marketing" fails. "Draft a reply to every new enquiry within an hour, using our service list, and flag anything about price" works.
The test for a good first job is simple. Could you explain it to a competent new contractor in two minutes and expect them to get it roughly right? If yes, it is a good first job. If explaining it takes twenty minutes and depends on things only you know, save it for month three.
Two: gather three real examples
This is the single highest-value thing on the list and the one everyone skips. Three real pieces of the work done well beat any description of how you want it done.
Go into your sent folder and find three replies you were happy with. Copy them out with the names removed. Saying "be friendly but professional" gives an AI Employee almost nothing, because those words mean something different to every business. Three of your own emails give it everything.
Three: write the one rule it must never break
Every business has a line. Never quote a price. Never promise a delivery date. Never say yes to a refund. Never mention a client by name. Write yours down before you start, not after something goes out.
Then check the platform can actually enforce it. Ask directly during your trial: what happens when a customer asks the forbidden question? A good AI Employee stops and flags you. A bad one answers helpfully and creates a problem you now have to unwind with a real person.
Four: list the tools it must reach
Walk through the job in your head and note every place you would click. Inbox, calendar, CRM, spreadsheet, phone, or a browser-only system with no proper connection.
That list is your real requirement, and it filters candidates faster than any feature comparison. Check each item specifically rather than trusting a wall of logos, because there is a big difference between reading data from a tool and being allowed to change or send from it.
Five: decide the number you will judge it by
Pick one measurable thing before you start, then write down today's value. Hours per week on the task. Response time to a new enquiry. Number of follow-ups actually sent.
Without a number, the review at the end of the month becomes a vibe, and vibes cancel subscriptions that were working. With a number, you either see the improvement or you learn something specific about why it did not happen.
| Item | What good looks like | Time to prepare |
|---|---|---|
| The job | One repeated task explainable in two minutes | 10 minutes |
| Three examples | Real past work, names removed | 10 minutes |
| The hard rule | One sentence starting with never | 5 minutes |
| Tool list | Every place you would click doing it manually | 10 minutes |
| The number | Today's value written down | 5 minutes |
A worked example: Tom and forty minutes of preparation
Tom is a mortgage broker working with one part-time administrator. His chosen job was document chasing, because clients routinely take three weeks to send payslips and every day of that is a day of his margin.
He pulled three of his own chasing emails, wrote one rule saying never discuss rates or eligibility, listed two tools which were his inbox and his case spreadsheet, and wrote down his number: average eighteen days from application to complete file.
Six weeks later the average was eleven days. Two things surprised him. The AI Employee flagged more cases than he expected, which turned out to be the rule working correctly. And the biggest gain was not speed of writing, it was that reminders went out on day three and day seven without anyone remembering to send them.
Red flags to watch for while you shop
A few signals reliably predict a frustrating first month, and they are visible before you pay anything.
- No way to test before paying: if you cannot run one real task without a sales call, you cannot evaluate it at all.
- No visible activity log: if you cannot see what it did and when, you are supervising on faith.
- No escalation setting: a platform with no concept of asking you first will eventually answer something it should not.
- Memory that is only chat history: if it cannot hold facts about a customer, you will re-explain your business forever.
- Invisible spend: if you cannot see cost accumulating and cap it, a heavy month becomes a surprise.
What your first AI employee will not do
It will not work well in week one. Week one is onboarding, exactly like a human hire, and expecting otherwise is the fastest way to give up too early.
It will not figure out what you meant. It will not notice that your prices changed unless you tell it. It will not take responsibility for a mistake, so anything customer-facing needs a rule and a review. And it will not fix a process that is broken on paper, because automating a bad process just produces bad outcomes faster.
Your first day, in order
- Hire one employee for the job you wrote — Match the role to the task. Do not hire three and split your attention across all of them.
- Paste the three examples first — Before any instructions. Let it see the work before it hears your opinions about the work.
- State the hard rule explicitly — One sentence, starting with never, plus what to do instead when the situation comes up.
- Connect one tool, not all of them — The one the job cannot happen without. Add the others once the first is behaving.
- Test the rule on purpose — Send it the forbidden question yourself. Confirm it stops and flags you before a real customer tries.
- Book fifteen minutes on Friday — Read what it did, correct two specific things, and check your number. That review is the whole management job.
Frequently asked questions
FAQ
What should my first AI employee's job be?
Something repeated, well-defined, and mildly annoying. Inbox triage, follow-up reminders, lead research, and document chasing are all strong first jobs. Avoid anything creative, high-stakes, or dependent on knowledge that only exists in your head, because those will fail and put you off the whole idea.
How long until my first AI employee is actually useful?
Expect the first week to cost you time rather than save it, because that week is onboarding. Most people see a real return somewhere in week two or three, once the brief has been corrected two or three times. If you have not seen anything by the end of week three, the job was probably too vague to start with.
Do I need to give it access to my email straight away?
Not necessarily, and starting narrow is wise. Many platforms let it draft without sending, so you can watch its judgment for a week before granting send rights. Once you have read twenty of its drafts and disagreed with none of them, widening access is an informed decision rather than a leap.
What if I do not have three examples of the work?
Then write one from scratch and use that. The point is showing a concrete finished piece rather than describing an abstract standard. If you genuinely cannot produce even one example of the job done well, that is a sign the job is not defined enough to hand to anyone yet, human or otherwise.
Should I tell my team I am hiring an AI employee?
Yes, and frame it as which task it owns rather than as a general capability. People cooperate with a clearly bounded colleague and resist a mysterious one. Being specific also stops someone quietly routing sensitive work to it that you never intended it to handle.
What does a first AI employee cost?
Start free wherever a real free tier exists, because you should prove the job works before you pay for it. Look at whether pricing follows seats or workload, since that decides what happens when you add a second employee. Sistava starts free with one AI Employee and no card, and paid plans start at 50 per month.