Name, face and job title
A colleague you can address rather than a nameless tool. It also makes delegation obvious once you have more than two of them.
Guide — — by Mahmoud Zalt
The whole hiring process in order: choose the role, interview, hire, onboard, brief, set approvals, run week one, review the work, and grow a team.
Start with the job, not the errand. The single most common mistake is hiring around one task, giving it one thing to do, and then judging the whole idea on a single output. A role covers a stretch of your week and keeps covering it. A task is a favour you asked for once.
The clean way to find the role is to look at where an answer stops being useful. Most AI tools give you a good answer and hand the rest back. You still open the CRM, create the records, write the emails, move the files, schedule the follow-ups, assign the work, review the result, and check nothing sensitive went out. That gap is the role.
Write it down as one sentence before you open the marketplace. "Someone who finds prospects, qualifies them, writes to them and chases the ones who go quiet." "Someone who answers support mail and only escalates the ones that need me." If you cannot write the sentence, you are not ready to hire, and no amount of setup will fix that.
You are choosing among 16 areas of a business, from sales and marketing to support, operations, HR and recruiting, finance and accounting, data and analytics, legal support, product, design and executive assistance. Over 150 live capabilities sit underneath those areas. The catalog is broad enough that the honest constraint is your clarity, not the range. The roles are listed in the open at Sistava, so it is worth reading the shortlist against your one sentence before you commit to any of them.
You can talk to a candidate before you hire it, and you should. It is a real conversation, not a preview: you ask questions, it answers, and you decide afterwards. This is the single most underused step in the whole process and the one that prevents almost every disappointing first week.
Interview it the way you would interview a person for the same job. Give it a situation from your actual business and listen for whether the answer is specific or decorative. Ask what it would need from you to do the work well, which is the question that separates a candidate that has thought about the job from one that is agreeing with you.
Hiring takes a minute, and what you set at this point shapes everything afterwards. Every employee gets a name, an avatar, a job title, a persona and a communication style, and none of that is decoration. It is how you and your customers will experience the work.
A colleague you can address rather than a nameless tool. It also makes delegation obvious once you have more than two of them.
Warm or clipped, formal or direct. Set it to match how your business actually sounds, because customers will read it.
How hard this employee thinks is a setting. Routine chasing stays light and cheap. Strategy, analysis and anything customer-facing gets more.
Active, paused, moved to the bench, or kept as a former team member with their history intact. Hiring is reversible without losing what was learned.
If the role you wrote in step one does not exist in the catalog, build it. The employee builder takes a job description in ordinary language and produces a custom employee shaped around it. Businesses are full of roles that have no standard title, and a custom one is a first-class hire rather than a workaround.
One warning before you move on. Do not hire five at once because the first interview went well. Hire one, run it through the rest of this guide, and let it be good before you add anyone. A workforce assembled in an afternoon is a workforce nobody briefed, and the review at the end of week one becomes unreadable.
Onboarding is the difference between generic output and output that sounds like your company. The employee learns from documents, websites, uploaded files, internal notes, past conversations, connected apps, databases, Notion and Google Drive. What you hand over in the first hour is what it will still be using in six months.
It keeps memory across runs, so context carries month to month. You explain your pricing structure once, not every Monday. It reads what you upload, including documents, so your existing contract template or price list becomes something it works from rather than something you retype into a chat box.
A job description here is not a document you file away. It is the working configuration of the role, and it is written in four separate layers so you can be strict where the process matters and loose where it does not.
| Layer | What it answers | Example |
|---|---|---|
| Duties | What this employee owns without being asked each time. | Owns the inbound support queue and the weekly pipeline review. |
| Skills | What it is able to use to do the work. | Research, writing, CRM work, spreadsheets, presentations, publishing. |
| Playbooks | The exact procedure and standard for a piece of work. | How a refund request is handled, in order, with the wording you approve. |
| Rules | How to behave in a specific situation. | Never quote a delivery date without checking stock. Always copy me on anything legal. |
The judgement call is how tightly to write the playbook. Strict is right for anything regulated, anything a customer sees repeatedly, and anything where a wrong step costs money. Loose is right for research, drafting and analysis, where fixing the path removes the value you hired for in the first place.
Now decide when the job happens. Working hours and a timezone are set per employee, which matters more than it sounds: it governs when messages land for the person receiving them, and it is the difference between a helpful follow-up and one that arrives at two in the morning.
Set two routines at most in week one. The temptation is to schedule everything you have ever wished happened automatically, and the result is a workforce producing output nobody reads. Two routines you genuinely read every week beat nine you learn to ignore.
Do this before you hand over work, not after something goes out that should not have. Control is what makes autonomy usable, and it is a five-minute job at the start compared to an awkward email afterwards.
Sending external email, publishing, spending money, deleting data, triggering a workflow, sharing confidential information, or anything over a threshold you set.
Approval can depend on the situation rather than one global switch. Routine and small goes ahead. Anything a customer reads waits for you.
Protection for sensitive data, policy enforcement, filtering of unsafe behaviour, reduced prompt-injection risk, and blocked information-boundary crossings.
Define what a good result must contain and must avoid. Output is checked before it reaches you and a failed check goes back for revision automatically.
Daily or monthly spending limits, credit monitoring, and simple work routed to efficient models while stronger ones are kept for work that needs them.
Dashboards, an activity timeline with a screenshot per action, work journals, cost tracking, an action inspector, and a view of exactly what your employee was told.
That last one deserves a sentence of its own. Being able to see the actual context an employee received before it acted is how you diagnose a bad output in thirty seconds instead of guessing. When something comes back wrong, it is usually not the reasoning, it is what the employee was working from.
One area is not left to your judgement. If you connect a Shopify store, every action that changes that store is approval-gated at the moment it runs, and that floor is enforced by the platform rather than being a toggle you could switch off. Reads and reporting run unattended. Other store platforms ask first on every single action.
Now hand over real work. The right first job is not the most impressive one, it is the one you have been postponing: the follow-ups you never sent, the CRM that has gone stale, the reviews nobody replied to. Choose something you can judge in a day.
A note on expectations for week one. A new hire is not at full speed on day three, and neither is this one. What you are looking for is a curve: fewer corrections on Thursday than on Tuesday, and a Friday output you would have sent yourself. If that curve is flat, the problem is almost always the onboarding pack in step four rather than the employee.
Review the artefacts, not the conversation. What comes back is real files you can open and send: Word documents, presentations, spreadsheets, PDFs, CSVs and images, with a document editor and a presentation builder in the workspace so nothing has to be exported to be usable.
It can edit an existing file rather than rebuilding it and destroying your formatting, which is what makes a second draft cheap. Combined with commenting in place, the review loop is short: mark the exact problem, get the corrected version, and move on rather than writing a paragraph explaining what you meant.
Everything lands in one workspace: a company drive, a content workspace, the built-in CRM, mailboxes, a company calendar, tasks, routines, projects, sprints and dashboards. There is one search across pages, employees, teams, tasks, projects, files, sprints and CRM contacts, and when you cannot find something you can ask your employee to take you there, opening the exact page instead of describing where it is.
Once the work is good, widen the surface. Every employee gets its own email address and can send and receive independently, so your support hire has a support address that customers can actually reply to. Connect Gmail or Outlook instead and it sends and replies from your real inbox under your name.
Beyond email there is web chat, Slack, Telegram and a personal mailbox, so briefing happens wherever you already work. Meetings are the step most people do not expect: add a Zoom, Google Meet or Teams call to the calendar, pick which employee attends, and it joins, listens, takes notes, can speak, and turns the conversation into follow-up work afterwards.
That last part closes a loop that leaks in almost every small company. The action item agreed on a call is normally remembered by whoever was least busy. Here it becomes a task attached to an owner while the call is still happening.
The second hire is easier than the first, and the fourth is where the shape starts to matter. At that point the question is no longer what each employee can do, it is who is coordinating them, and the answer should not be you doing it manually every morning.
Give the team an AI leader. The leader delegates to the members and reviews what comes back before it reaches you, which is the difference between managing four employees and managing one relationship. You can also hire a complete team in one go rather than assembling it hire by hire.
Set team objectives, goals and guidelines so the workforce is aimed at an outcome rather than a queue of requests.
A real structure for your workforce, so who reports to whom is visible instead of remembered.
A room generated from your real teams, where each employee is a character at a desk you can watch, click and message live.
A goal, a weekly rhythm, a retrospective and a written account of what was actually completed.
There is also an AI mentor watching the whole workforce. It chases stalled work, resolves the common blockers, retries where retrying makes sense, and escalates only when a human decision is genuinely required. That is what keeps a five-person workforce from quietly turning into five things you have to check.
Reach is rarely the constraint. There are 874 connected apps and services, plus web search, website reading, browser automation, computer control, screen vision, terminal commands, file organisation, data-entry automation and meeting transcription. With your permission an employee can operate the same browser and desktop applications you already use, through a companion app for macOS, Windows and Linux, so a tool does not need a perfect connection to be reachable.
The eleven steps above would be a chore to configure by hand, which is why you do not have to. The personal assistant at Sistava is the control layer for the platform: it sets up the workspace, understands your company, hires and configures employees, creates projects and tasks, connects apps, establishes routines, finds files and contacts, navigates you anywhere, reviews activity, checks progress and spending, and adjusts how the workforce operates.
It will also explain the platform itself when you are unsure what something does, which means every step in this guide has a shortcut: ask for it. "Set working hours for my support hire to nine to five in my timezone and require my approval on outbound email." That is the configuration, spoken.
| Dimension | Traditional | With Sista |
|---|---|---|
| Choosing | You pick a tool and then work out what it is for. | You write the role in one sentence, then hire for it. |
| Evaluating | You read a feature page and hope. | You interview the candidate and judge the answers yourself. |
| Briefing | You re-explain the context in every new conversation. | Duties, playbooks and rules are written once and memory carries forward. |
| Controlling | You either trust it fully or check everything by hand. | Approval rules, output checks and budget limits set per situation. |
| Reviewing | You scroll a chat log trying to reconstruct what happened. | An activity record with screenshots, a work journal and a cost figure. |
| Scaling | Every extra tool is another thing you personally operate. | A team with a leader who delegates and reviews before it reaches you. |
That is the whole loop. A chatbot ends at the answer, while a workforce moves a request through research, planning, execution, communication, review, storage, measurement and follow-up. You chat, they work, and the hiring process above is what makes the difference between those two outcomes.
Minutes to hire, a week to get good. Choosing a role and completing the hire takes a few minutes, and you can hand over a real job the same day. The onboarding, the playbooks and the approval rules are what turn a decent first output into a reliable one, and that work spreads across the first week. Treat day one as calibration rather than delegation and the curve is much faster.
Yes. You can interview a candidate before hiring, which is a genuine two-way conversation rather than a demo. Ask it to describe the role back to you, give it a hard situation from your own business, and ask what it would need from you to start. There is also a free tier with lifetime credits, so you can put a real job in front of a hire before making a decision. Plan details are on the pricing page.
Five things: what the business does and for whom, your pricing and commercial boundaries, three real emails that show your tone, your policy and escalation rules, and access to the sources the role genuinely needs. It learns from documents, websites, uploaded files, notes, past conversations, connected apps, databases, Notion and Google Drive, and it keeps that context across runs so you explain it once rather than weekly.
Set approvals before the first task. Human approval can be required on sending external email, publishing, spending money, deleting data, triggering a workflow, sharing confidential information, or anything over a threshold you choose, and the rules can be written in plain language so approval depends on the situation. You can also define what a good result must contain and must avoid, and failed output is sent back for revision before it reaches you.
Only if its role permits it. Memory and knowledge follow the role, so an employee that may not reach a resource cannot get at that information through memory, through a knowledge search, or by asking in a different conversation, and what it writes back to memory is bounded the same way. Change the role and the access changes with it. The platform enforces this rather than relying on the model to keep a secret.
One. Hire a single role, run it through onboarding, playbooks, approvals and a first week, and let it get genuinely good before adding anyone. Hiring several at once produces a workforce nobody briefed and a review you cannot read. Once you have two or three working well, give the team a leader who delegates and reviews what comes back, and add OKRs so the group is aimed at an outcome.
Employees have a status lifecycle. You can keep one active, pause it while a season is quiet, move it to the bench, or keep it as a former team member with its history intact. Nothing you taught it is destroyed by pausing it, so bringing it back later does not mean onboarding it again. You can also set daily or monthly spending limits so a dormant period never becomes an expensive one.
Read the artefacts and the journal instead of the conversation. Each employee keeps a work journal of what it did, what it decided and what it hit, and the activity timeline records every action with a screenshot. When something looks wrong, an action inspector and a view of exactly what the employee was told will usually show the cause in seconds, which is almost always the context it was working from rather than its reasoning.
If you follow only three of the eleven steps, make them the interview, the onboarding pack and the approval rules. Those three take under an hour between them and they decide almost everything about whether week one feels like a hire or a gadget. All three sit in the first session after you sign up at Sistava, so the hour is spent once and every week after it is review rather than setup.
The rest is the same management you already do: give someone a clear job, tell them how you want it done, let them work, read what comes back, and correct it once so it never needs correcting again. What you get back is the stretch of the week that used to sit after every answer, and that stretch is where your evenings went.