The most expensive obstacle for an AI project is not the budget and not the integrations. It is the phrase "they want to replace us," muttered by the water cooler. Nobody says it out loud in meetings. Instead, the data for the prototype takes weeks to hand over, every mistake the agent makes is discussed louder than any success, and at acceptance testing ten reasons suddenly appear for why "this won't work for us." We see it regularly in our reviews: pilots die not from the technology but from quiet sabotage. That is why a conversation about what really happens to jobs is not sentimentality but part of the economics of adoption.
Why "replacing people" doesn't add up, even in theory
Companies don't come to us with the task of "cutting headcount." They come with the task of "our people are drowning in busywork, and there's no one to hire." A qualified specialist is scarce in Kazakhstan: a clinic spends months looking for a doctor, a strong sales manager takes a long search to land, and recruiting and onboarding a single office employee costs 1–2 million ₸ once you count the manager's time and the first unproductive months.
Firing a person you spent that kind of money to find, just to "replace them with a model," is economics with a minus sign. The agent takes over operations, not the position: filling in records, moving data between systems, drafting documents, standard replies. Decisions, accountability, and exceptions stay with the person. The platform is designed so that irreversible actions pass through control points where a human confirms or rejects the result.
The doctor: two fewer hours of paperwork a day
An appointment lasts fifteen minutes, and for a third of that time the doctor is looking at a form rather than at the patient. After the shift comes another hour to hour and a half of filling in charts, referrals, and codes. In total, about two hours of documentation work a day.
With the platform, the doctor dictates the examination by voice: the agent structures the entry into the chart, suggests ICD codes, assembles the case, and checks it for completeness before submission to the fund. The doctor reads and confirms. Two hours go back into the appointment: that's two or three extra patients per shift, or a normal lunch break the doctor hasn't had in years. No chief physician cuts doctors in this setup: there aren't enough of them as it is. We broke down how this works in the solution for medicine.
The sales rep: CRM stops being a punishment
The rep lives in two worlds: in conversations with clients and in the records the manager demands. Records eat 1.5–2 hours a day, they get filled in at night from memory, so only a third of reality makes it into the CRM. The manager knows this and demands more strictly; the rep hates the system even more. Sound familiar?
The agent processes every call right after the conversation: it fills in the record itself, logs the agreements, sets the next step. The rep checks it in two minutes. Both sides win: the rep gets two hours for selling, the manager gets a pipeline they can trust. At this point companies grow not their headcount but their output: more deals from the same person. The mechanics are in the solution for the sales department.
The accountant: from data-entry operator to controller
The peak days of month-end close mean hundreds of acts, invoices, and delivery notes keyed in by hand, up to three days of solid data entry. The agent recognizes source documents, including handwritten ones, posts them into the accounting system, and highlights discrepancies with contracts and statements. The accountant is left with the exceptions: the places where real expertise is actually needed. The role shifts from data entry to control and methodology: less mechanics, more substance. This is exactly what the people who chose the profession expected from it.
What actually changes in teams
The structure of the day changes. Instead of "doing it by hand" it becomes "checking and deciding": the employee approves the agent's results at control points, handles exceptions, labels errors. A new skill appears that quickly becomes valuable on the job market: the ability to formulate the rules of a process so that an agent can carry them out.
The fear of the "black box" is cured by transparency. Every step the agent takes is written to an activity log: who started it, what it read, what it changed, who confirmed it. When the team sees this log on day one, the conversation about "they'll replace us" turns into a conversation about "I have the final say." We described how the log works and the access model on the security page.
How to roll it out so the team doesn't resist
A short list from our rollout experience:
- name the scenarios by the busywork, not by job titles: "processing source documents," not "automating the accountant";
- show the activity log and control points on day one, not when someone asks;
- set the process result as the metric – speed, completeness, quality – not "payroll savings";
- give the team the job of labeling the agent's errors: whoever teaches stops seeing the student as a competitor.
The typical dynamic we see in support: for the first two weeks operators pointedly double-check every answer after the agent. By the end of the second month, standard questions go to the agent entirely, people work on the complex cases – and it is the complex cases, not a conveyor belt of identical replies, that keep strong operators from quitting.
Frequently asked questions
So there are never any layoffs at all?
The honest answer: reallocation does happen. In practice, companies more often freeze planned hiring than let go of current staff: the busywork grows faster than the headcount, and the agent closes that gap. The typical picture a year later is the same team handling a larger volume of work.
How long do employees take to get used to the agent?
Two to four weeks with a proper rollout: control points from day one, a clear log, the right to reject the agent's result. The free prototype we set up in about a week helps here too: the team sees the agent on their own examples before it appears in the workflow.
Who is accountable if the agent makes a mistake?
The process owner, just as with an employee's mistake. The difference is that with the agent every step is recorded in the log, the error is reproducible and fixed with a rule once, and irreversible actions are simply not available to the agent without human confirmation.