Job postings show what businesses expect from AI specialists: what tasks they assign, what skills they require, and how much they offer to pay. We analyzed 230 listings to help managers decide who to hire and help candidates understand requirements and choose the right role.
The sample includes job postings from hh.ru for AI implementation, development, project management, analytics, and training. An agent collects them from search results, applications, and saved listings, preserving full descriptions. Duplicates and vacancies without AI-related tasks are excluded from the count.
Data was collected on October 21, 2026. The sample reflects the tasks and requirements found in the collected postings. Assessing demand growth across the wider market requires regular measurements with consistent queries: the number of saved listings also depends on our selection criteria.
This breakdown is updated when new postings change figures or conclusions. The last update date is in the header. Since the previous update on October 1, 65 postings were added to the database.
A job title does not explain who is being hired
The title “AI Engineer” covers noticeably different responsibilities. In one posting, it means developing models and working with data; in another, automating procurement, logistics, or customer service. Consider the second first: this is where it is easiest to make a mistake by judging candidates solely by past job titles.
The job posting for a Senior AI Engineer at Internet Revolution lists product and supplier search, profitability calculations, content preparation, advertising, and logistics. The engineer must understand what employees do manually and automate that work. For example, build a system to find promising products and calculate shipping costs. The company expects both development skills and an understanding of how the business makes money.
Such a role is difficult to classify purely as engineering or purely as management. The person will have to discuss tasks with procurement specialists, work with data, and write code. If a company hires a strong developer and expects someone else to handle procurement, critical responsibilities will remain uncovered.
Here are the responsibilities associated with different roles in the sample.
| Role | Assigned tasks |
|---|---|
| Implementation and automation specialist | Identify manual work, automate it, integrate required systems, and train employees |
| AI engineer | Develop agents or document search, configure models, and maintain them |
| Practice lead | Select projects, justify budgets, organize work, and measure impact |
| Project manager | Gather requirements, agree on deadlines, coordinate development and rollout |
| Product manager | Validate who needs the AI product, select features, and track usage results |
| Analyst | Examine data and workflows, test hypotheses, and prepare requirements for development |
| Training specialist | Prepare training sessions and guides, help employees use AI in their daily work |
For instance, the project manager at Exon Group manages a team consisting of a business analyst, developers, and machine learning engineers. This person gathers requirements and plans development. That is different from searching for suppliers and writing procurement automations independently, even though both roles involve AI implementation.
Companies ask to connect AI to internal data and software
Job descriptions frequently mention APIs, internal accounting systems, and knowledge bases. Behind these requirements lies a clear task: retrieve company data, process it with AI, and deliver the result back where the employee works.
The chart accounts for direct mentions, including skills listed as an added advantage.
APIs allow applications to exchange data. RAG retrieves information from company documents for the model to reference in its response. An agent can perform multiple actions in sequence: find a document, query another application, and prepare the result.
For example, preparing a response to a customer inquiry might require support ticket history from CRM, contract terms from a document repository, and stock balances from the inventory system. A specialist must understand where to get this data and how to verify the generated response. Being able to use a chat interface is not enough: integration with data and company software is required.
Automation platforms like n8n appear alongside Python and SQL. A procurement job posting from Sber describes this division directly: the specialist quickly builds and validates a prototype, and the product team later develops a production version. They are hiring someone who can test an idea before significant engineering resources are allocated to it.
Some roles require on-premise servers and fine-tuning
Not all employers plan to use cloud services. The AI/LLM Engineer posting at Russian Railways explicitly bans external APIs: models, databases, and data processing must run entirely inside closed infrastructure. The candidate must have experience deploying local models and integrating them with internal systems.
Local models or on-premise deployment appear in 22 descriptions; fine-tuning appears in 15. Among the 65 postings added since October 1, these requirements and preferences appear in 6 and 4 listings, respectively.
At Systeme Electric, the implementation specialist’s responsibilities include deploying and fine-tuning models on a server. Fine-tuning adapts an off-the-shelf model to a specific task using additional examples. The same posting requires coordinating solutions with information security; experience with GPUs and inference acceleration is listed as an advantage.
At Mebel Rus, fine-tuning is tied to a specific task. The company wants to generate designs in Bazis-Mebelshchik (furniture design software) from measurement photos: recognizing walls, windows, doors, and dimensions, then generating rooms and furniture. This requires preparing training data, processing images, and fine-tuning a model. Experience in furniture design is an advantage.
In these postings, technical requirements stem directly from the task. Sensitive data requires internal infrastructure. Measurement recognition requires sample images and testing on actual rooms. Employers care whether candidates have built similar solutions and understand what it takes to launch.
Quality evaluation can be a standalone specialization
HeadHunter is hiring an engineer for its LLM Evals team. The job involves building benchmark evaluation datasets, configuring automated evaluation for model responses, establishing metrics, and analyzing assistant errors. Here, “evals” refers to quality assessments of language model solutions.
At Meteoro Platform, this responsibility belongs to the engineer building the agents. The description offers a solid criterion: any discovered failure must become a test. If an agent fails on a specific task, that task is added to the test suite to catch regressions after future changes.
At National Transport Company, acceptance is also tied to evaluation datasets built on real-world data. Evaluation becomes part of the assignment and acceptance criteria. A job posting like this allows candidates to see in advance how their work will be evaluated.
For managers, the key takeaway is how the task is structured. Who will collect test cases, review outputs, and analyze errors post-launch? The surveyed postings show both options: a dedicated specialist or an engineer responsible for both development and evaluation.
Domain experience in procurement or document workflows is also becoming a requirement
A procurement posting at Sber requires at least two years of experience in procurement, tenders, or supply chain. The future employee must understand procedures and work with technical specifications, contracts, and commercial proposals. AI tools will be used to automate this work. Procurement expertise is listed as a mandatory requirement alongside experience building working AI solutions.
A posting for a document workflow architect offers two acceptable background options: the candidate has already automated contract, HR, or accounting workflows, or did something similar in another domain and is ready to learn document processes. A law degree is not required. The employer asks candidates to showcase a system they built and compare workflows before and after automation.
Among the 59 job postings added in September, there are five analyst roles. They involve working with data, testing hypotheses, gathering requirements, and evaluating model outputs. For example, an AI analyst role at Strana Development involves testing prompts and improving outputs. This is also AI implementation work, with its own specific candidate requirements.
Professionals with experience in procurement, document workflows, or analytics have good reason to consider these roles. Both sides of the requirements matter: deep domain knowledge and proven automation skills.
Some employers define expected results for the first few months upfront
At Home-Heat, the head of AI implementation is given a 90-day target: identify priority processes and deliver two to three projects with measurable impact. Applicants are asked to describe past projects with quantified results.
In Sber’s procurement role, impact is measured in hours saved, cost reductions, or error decreases. The document workflow architect listing specifies document preparation time, overdue document rates, and verified cost savings. These criteria appear both for management roles and hands-on automation specialists.
To meet these expectations, candidates need to know how to measure current workflows: how many documents are processed monthly, how long processing takes, and how many errors need fixing. After rollout, these baselines can be compared against the new process while accounting for tooling and maintenance costs. More on this in the breakdown of implementation economics.
The number of deployed agents does not replace this comparison. In postings that define results in detail, employers care about how much manual work was eliminated and what actually changed for employees.
Employee training can be a dedicated role
Severstal is hiring a generative AI and employee training expert. The role involves creating courses and practical workshops, training staff of varying skill levels, and helping them apply AI tools to day-to-day tasks. Impact is evaluated by whether employees continue using AI after training and how much time it saves.
At Systeme Electric, training is part of the implementation specialist’s responsibilities. After launch, they must prepare documentation, train the client, gather feedback, and support the product. The employee handles the entire scope end to end.
This is another distinction worth clarifying before hiring. Will the engineer focus strictly on development, or will they also teach staff, write guides, and answer user questions? Job postings show both approaches.
Managing employee resistance is written into job descriptions
The word “resistance” appears directly in 5 out of 230 postings, while training employees or users is mentioned in 49 out of 224 full descriptions—roughly one in five. The rationale is evident in the listings: a tool that employees fear or avoid produces no results. Someone must train staff and address their objections.
This is described in greatest detail by OMEGA in a lead AI transformation specialist role paying ₽400k before tax. The responsibilities feature a dedicated section: “Training and managing resistance.” The specialist must “conduct practical workshops solving real employee tasks rather than lectures” and “actively handle staff objections and fears, turning them from AI opponents into allies.” Directly following this section are metrics: time and cost savings tracked for each process.
In other postings, this appears as a one-line task, but the wording is just as direct. An AI Business Analyst at Novaroll is tasked with “training people. Demonstrate, explain, remove resistance—so that the tool is actually used.” Rosgosstrakh expects an AI automation lead to “share results, resolve resistance, and build trust in the practice.” At Agrotek and Solid Financial House, handling resistance falls under change management—preparing people for a new way of working.
Managers should decide this before hiring: who will train staff and address objections—the new hire, department heads, or someone else? Candidates should prepare examples of how they helped colleagues adopt new tools: in some postings, this work is listed among the duties alongside the technical work.
Salaries are listed in roughly a quarter of job postings
Out of 230 postings, salary is specified in 63—roughly 27%. The remaining 167 do not disclose a figure.
Here are three examples. Figures are take-home.
| Position | Listed compensation |
|---|---|
| AI Implementation Specialist | ₽50–60k |
| AI Engineer for Local Models | ₽150–250k |
| AI Project Manager | ₽200–300k |
The first position has an existing prototype of an AI assistant for doctors and a technical specification. The task is to refine search across verified medical literature and chat functionality; domain experts will review the content. The second involves running models on internal hardware and integrating other services. The third requires managing a development team on a construction tech product.
The pay gap reflects the difference in scope. These three examples illustrate the range of offers, but they cannot establish a market average. Other postings also include figures before tax and in different currencies; they are not normalized here.
How to use this analysis
For managers, these job postings help define the scope of work before searching for a candidate. If procurement needs automation, consider who understands the procedures, who can build the solution, and who will review documents. A single hire might combine these duties, but domain experience then becomes just as essential as technical skills.
During interviews, this is best evaluated through a single project from the candidate: what was done manually, what they changed themselves, how they verified results, and who used the system after launch. Discussing a concrete project grounds the conversation in the exact work the candidate will be expected to perform.
Candidates should read listings with the same questions in mind. Which tasks will they handle alone? Is there a team? Who will supply the data and verify the results? Is knowledge of procurement, healthcare, or manufacturing required? This makes it easier to see if the position fits their background and what they will need to learn.
The title “AI specialist” still encompasses too many disparate responsibilities. Postings that clearly outline the task, the team, and expected outcomes provide far better grounds for decisions—for both employers and applicants.