Will AI take our jobs? Facts and forecasts

AI does not “take jobs” in one move, but gradually transforms what the working day looks like across many professions. It most often automates repetitive tasks: preparing first drafts of content, organising data or handling standard queries, which increases the importance of quality control and accountability. This raises concrete questions about full-time roles, salaries and whether entering the market (especially at junior level) will become more difficult. At the same time, implementations in companies usually take months, because they require integration, testing and compliance approval, so changes come in waves. In this article, we separate the mechanisms of automation from media headlines and show how to read the data and market signals. To assess your own risk without panic and without illusions, it is worth starting by understanding what AI can actually do, and what it still cannot be “handed over” to it.

How AI actually automates work: mechanisms and limitations

AI most often automates tasks, not whole professions, so a single role is usually “redesigned” rather than simply disappearing. In practice, models take over specific activities such as creating first drafts of texts, summarising, organising data, classifying tickets or generating boilerplate code. The effect is often that an employee spends less time on routine work and more on decision-making, client contact and quality control. The risk to an employee depends primarily on what percentage of the day is taken up by repetitive tasks, not on the job title itself.

Large language models (e.g. ChatGPT, Claude, Gemini) work like “language engines” and are best at creating and transforming content: emails, reports, proposals or FAQs. At the same time, they do not guarantee truth, because they can hallucinate, get facts wrong, cite non-existent sources or lose the company context. For this reason, organisations most often use them for draft versions rather than for making final decisions without verification. As a result, the importance of checking assumptions and adapting responses to the realities of the process increases.

To ensure AI responds in line with an organisation’s documents, companies use Retrieval-Augmented Generation (RAG), i.e. combining LLMs with searching a knowledge base. For example, an HR chatbot can use policies and procedures in SharePoint or Confluence instead of the internet’s “general knowledge”. This improves accuracy, but requires orderly documentation and ongoing maintenance of the knowledge index. Without this, response quality drops and automation falls back to the prototype stage.

Process automation in companies often relies on RPA (e.g. UiPath, Automation Anywhere) and low-code tools such as Power Automate or Zapier, which connect systems and remove repetitive tasks from people’s work. Such workflows can transfer data, generate PDFs, update CRM and send emails, while AI brings “intelligence” to areas where human work was previously essential (e.g. email classification, extracting fields from contracts or recognising content from invoices). As a result, one person can handle a larger number of cases without a proportionate increase in working time. The biggest gains go to teams that operate on measurable processes and have structured input data.

• Generating and rewriting content and summaries (LLMs), but most often as draft versions for verification.
• Handling flows between systems (RPA/low-code), e.g. updating records and sending communications.
• Supporting developers with boilerplate, tests and refactoring (GitHub Copilot, Codeium), with an increasing role for code review and security.
• Facilitating data analysis (AutoML, BI with natural language), with a constant need to understand metrics and data quality.

The limitations of automation arise primarily from accountability, law and reputational risk, because someone has to “sign off” the result. In many industries, it is not possible to hand decision-making over to AI completely: a doctor, accountant, lawyer, auditor or project manager are still responsible for the consequences. That is why companies adopt a human-in-the-loop approach and require auditability, which usually shortens working time rather than leading to the elimination of jobs. In addition, many projects are held back by bottlenecks: data silos, lack of APIs, file chaos and GDPR constraints, as well as increasing cybersecurity requirements.

1. 01Language engineCreating and transforming content (emails, reports)
2. 02Task automationTaking over routine tasks, organising data
3. 03Redesigning the roleLess time on routine work, more on strategy
4. 04Human decision and controlClient contact, decision-making, quality oversight

Key takeaway: AI does not replace roles, but changes their structure, enabling focus on higher value-added tasks.

Facts and figures: what research says about AI replacing work

Studies and reports more often describe shifts in tasks and skills than an immediate “end of work” caused by AI. The World Economic Forum emphasises that technology is simultaneously phasing out some roles and creating others, and what matters is how companies reshape skills. The OECD points out that a high share of automatable tasks affects a minority of workers, but for a large share there is a significant component that can be automated. This explains why the pressure is felt so widely, even though professions do not disappear overnight.

Loud forecasts about “hundreds of millions of jobs” most often refer to exposure to automation of parts of tasks, not the complete replacement of full-time roles. A model can generate a document, but someone has to adapt it to the risk, conduct negotiations and answer to the client or the court, so context and accountability remain key. It is worth reading the methodology of forecasts carefully: the devil is in the detail, i.e. whether we are talking about hours of work or positions, and what the time horizon is. This approach makes it easier to distinguish real operational changes from headlines alone.

In Poland, areas of office work connected with shared services (BPO/SSC) are particularly susceptible to automation, as well as administration and document handling, where the influence of RPA and generative AI is strongly felt. In industry, the pressure covers quality control, planning and maintenance, where the importance of vision systems and fault prediction is growing. At the same time, demand is increasing for people who can implement tools and keep processes compliant with the law. In practice, this shifts the market towards digital skills, data work and security.

The pace of change in the labour market is slowed by the fact that implementations take months, not days, because they require integration, testing, procedures, training and compliance approvals. The differences between sectors are clear: in banking, insurance and medicine, automation is often slower due to audit, explainability and data protection requirements, whereas in marketing, media and e-commerce the pace is faster. From an HR perspective, the phrase “will take jobs” often means limiting recruitment for junior roles, combining roles, or raising requirements (e.g. automations and prompting within responsibilities). The most reliable signal of change remains how tasks and requirements evolve in job adverts, rather than general forecasts themselves.

Occupations and tasks most exposed to automation

The most exposed to automation are occupations in which a significant part of the day boils down to a repetitive “information flow” between systems and documents. This includes, among other things, data entry and back-office work based on transcription, validation and forwarding of records, which can be automated using RPA and document data extraction (OCR + NLP). In practice, people are increasingly dealing with exceptions, while routines are taken over by workflows. The key question is: what proportion of the work is routine document handling, and what proportion is interpretation and decision-making.

In first-line customer service, the risk mainly concerns standard, repetitive enquiries that can be handled by chatbots and voicebots. Typical examples include parcel status, password resets, simple complaints or product information, and the human role shifts towards escalation and non-standard cases. The scale of change depends on whether the organisation measures the share of cases “closed” without an adviser (deflection rate) and how quickly self-service effectiveness grows. As a result, pressure does not have to mean immediate redundancies, but it often translates into fewer full-time roles in the first line as automation matures.

Roles based on “volume” content production and translation of uncomplicated materials are also highly exposed. Translation tools handle instructions, product descriptions and correspondence efficiently, while generative AI lowers the cost of preparing basic SEO texts and mass copywriting. At the same time, tasks that require legal responsibility, precise sector terminology, consistency and quality control remain better protected, where the importance of the editor and the person safeguarding standards is growing. From a market perspective, this translates into price pressure where content does not require unique know-how or rigorous research.

Automation also covers tasks with repetitive, easy-to-reproduce scenarios, such as manual testing of stable flows in web applications or cyclical financial reporting in “copy-paste” mode. In IT, designing test strategies, working with risks, test data and test automation (Playwright, Cypress, Selenium) is becoming increasingly important, rather than manually repeating the same cases over and over. In finance, monthly summaries and repetitive reports are easier to automate using BI, ETL and generating descriptions for charts than actions requiring interpretation of scenarios and recommendations for stakeholders. A similar mechanism can be seen in mass recruitment, where CV screening and first-contact communication can be supported by AI, although decisions often remain with humans due to the risk of error and regulatory requirements (RODO, labour law).

1. 01Information flowTransfer between systems
2. 02Data entry & back-officeTranscription, record validation
3. 03Exception handlingA human makes decisions
4. 04Standard serviceSimple, repetitive enquiries
5. 05Automating routine workWorkflows take over tasks

The key is the proportion of routine handling to work requiring interpretation and decision-making.

Occupations and skills growing thanks to AI

Occupations and skills growing thanks to AI are above all those that help companies organise data, carry out safe implementations and turn technology into measurable processes. Demand for data engineering and integrations (ETL, API, lakehouse) is clearly growing, because the real value from AI appears only when data is clean, accessible and linked to the organisation’s operations. In practice, tools and technologies such as dbt, Airflow, Spark, Snowflake and BigQuery are most often mentioned, as well as API integrations and queues (Kafka). This is an area of work that often determines whether automation moves beyond the prototype stage.

Cybersecurity and data protection are also growing strongly in importance, because AI simultaneously increases risks (e.g. generated phishing) and supports defence (log analysis, anomaly detection). As a result, roles such as SOC analyst, cloud security and GRC are strengthening, along with competencies related to data classification and DLP. In many organisations, it is security that decides whether tools such as ChatGPT/Copilot can be used on sensitive data. If a company cannot manage data and access securely, AI implementations will be limited regardless of the tools’ potential.

At the intersection of business and technology, roles related to product management and designing processes with AI are growing, because someone has to translate the capabilities of tools into KPIs, confidence thresholds and exception escalation rules. In parallel, the area of MLOps/LLMOps is developing, where implementation and maintenance of solutions become key: monitoring drift, inference costs, response quality and compliance with policies. In practice, tools and approaches such as MLflow, Weights & Biases, Docker/Kubernetes are becoming widespread, and in the world of LLMs, prompt control, logging, guardrails and response regression tests are also important. These are strictly production competencies, not a one-off “model launch”.

The growth in demand also includes AI-augmented domain specialists and people who can build and maintain automations within an organisation. Experts (e.g. lawyers, doctors, engineers) gain productivity thanks to support in research, document creation and case analysis, but professional responsibility and decision-making remain with the human. At the same time, the importance of creators and operators of automations (Power Platform, UiPath), the so-called citizen developers, is growing as they connect tools and remove manual, repetitive steps from operational processes. Companies are also strengthening business analytics and data storytelling skills as well as internal education (AI literacy), and as regulation expands, demand is growing for compliance, audit and AI law (EU AI Act, GDPR), with an emphasis on documentation and risk management.

How work in companies will change: organisation, KPI, working style

Work in companies will change primarily through faster pace and higher productivity standards, because AI shortens the time needed for many tasks and raises expectations around the number of iterations and response speed. In practice, this does not always mean reductions, but more often translates into a greater number of projects per person and a shift towards more complex tasks. For many teams, the key question will be whether the organisation reinvests the recovered time into development or treats it solely as an opportunity to cut costs. The change will be visible both in operations and in creative and analytical work.

The junior–mid–senior pathways will be reshaped, because AI is taking over some typically junior tasks, such as the first draft or simple analyses. Companies may hire fewer juniors or expect more from them from day one, which makes it harder to “enter the profession”. In such conditions, internships, portfolios and self-initiated projects become more important as proof of skills. At the same time, hybrid work and distributed teams may run more smoothly thanks to automatic notes and summaries, which changes outsourcing calculations and the way work is organised.

Quality will become just as important as speed, because organisations will measure more carefully the errors and consequences of AI “hallucinations” in processes. Companies are implementing checklists, tests, peer review, source validation and acceptance thresholds, and KPIs are increasingly including, among other things, the share of cases requiring correction, the number of escalations and complaints after automation. The ability to spot AI errors and work according to clear quality criteria will be a core competence for many roles, not an “option”. This shifts the work culture from “production” to verification and decision-making.

Teams will also need new operational roles, such as automation owner and knowledge steward, because prompts, knowledge bases, integrations and escalation rules require ongoing maintenance. Without assigned responsibility, solutions degrade over time: answers become inconsistent, the number of exceptions grows, and users lose trust. In parallel, AI encourages a reorganisation “from functions to value streams”, where end-to-end teams with shared KPIs gain importance instead of tasks being handed over along a chain. An additional part of everyday work will be controlling tool costs (tokens, licences, cloud) and optimising usage, because the budget is linked to the intensity of use.

Competitive advantage will be built by organisations that treat knowledge management as the foundation of AI implementations. When procedures, glossaries and the decision base are organised, AI can function as a scalable assistant for employees. When knowledge remains “in people’s heads” and in chats, answers will diverge from one another and may deepen the chaos. That is why the role of repositories and the single source of truth approach is growing, as they stabilise quality and make process audits easier.

1. 01Faster pace and higher standardsFaster iterations, higher expectations
2. 02Task complexityMore projects, higher complexity
3. 03Reinvestment of timeDevelopment vs. cost cutting
4. 04Restructuring career pathsFewer juniors, AI takes over simple tasks

AI speeds up work and changes the structure; the key is to reinvest the recovered time into development.

How to protect yourself as an employee: practical strategies

You can best protect yourself as an employee when you consciously limit the share of routine in your work and take on elements of responsibility, quality and influence over decisions. It is worth starting with an audit of your own tasks using the 30/50/20 method: divide the week into routine tasks (easy to describe), analytical tasks (require context) and relational tasks (require trust). If routine accounts for more than 50%, this is a sign of high exposure and a need to quickly strengthen your tool and domain skills. This approach gives a more concrete picture than asking “will my profession disappear”.

The most “AI-proof” value is the one based on responsibility and decisions, because AI does not take over the real signature under risk. This applies, among other things, to negotiations, risk assessment, prioritisation, client relationships and the role of KPI owner or quality lead. It is worth planning to take over at least one area of responsibility in the team, because this shifts you from an execution role to a role that is essential to the process. In practice, AI can speed up the preparation of materials, but responsibility for the outcome remains with the human.

The biggest return over the 6–18 month horizon comes from learning “here and now” tools and building habits of working with quality, because this quickly translates into visible improvements in the process. In many roles, mastering just 2–3 solutions is enough to genuinely shorten day-to-day work, such as drafting, meeting summaries or automating file and email workflows. It is worth treating prompting like a communication skill: clarify the goal, the model’s role, the context, constraints and the output format, then force verification by asking for sources, comparisons and a list of assumptions. In this approach, AI works like an “intern” you guide and supervise, rather than an oracle.

• Document the “before/after” impact (process time, number of errors, NPS, number of tickets per person), because this is the strongest argument in the market and in the company.
• Choose a domain specialisation instead of being generalist (that is where AI “eats” jobs more slowly), combining it with tools and automations.
• Build a foundation in working with data (advanced Excel/Sheets, Power Query, SQL basics, reading BI dashboards) so you can verify results and talk in numbers.
• Become a “multiplier” in the team: standardise practices (prompt templates, quality checklists, automation libraries) and help others increase productivity.
• Treat security and ethics as a competence: use anonymisation, work on summaries and follow confidentiality rules, instead of pasting sensitive data into public tools.

Your professional position can also be stronger when you can combine tools with a safe way of working in the organisation. Knowledge of the basics of GDPR, confidentiality rules and data leak risks helps avoid missteps that block rollouts or undermine trust in automation. At the same time, it is worth developing a portfolio of results, because documented time savings, quality improvements or sales growth are more convincing than declarations of skill alone. In practice, it is measurable impact and the ability to maintain quality that distinguish people “augmented” by AI from those pushed out by automation.

What employers and the state should do to avoid mass redundancies

Employers and the state can reduce the risk of mass redundancies if they treat AI rollouts as a transformation of tasks and processes, rather than as a one-off cost-cutting exercise. The best results come from on-the-job reskilling programmes, meaning those that combine learning with a real project: automating a process, implementing a bot or cleaning up data. Training on its own, without implementation, does not change how a company operates, so it does not reduce the risk to jobs. The key is assigning responsibility: who will implement the solution, measure the effect and maintain it.

To ensure automation does not undermine quality and reputation, organisations should implement a “human-in-the-loop” policy and clear decision thresholds. In practice, this means defining where AI can act autonomously and where it can only support a decision (for example, prepare a reply, but sending it requires consultant approval until the error rate falls below an agreed threshold). This model reduces legal and reputational risk, while also allowing productivity to rise without abruptly “cutting out” roles. This approach also makes it easier to redesign roles so that people move from routine work to tasks of higher value.

A fair transformation also depends on standardising data and processes, because process chaos makes sensible workforce shifts difficult. Process mapping (BPMN), KPI definitions and data owners help move people from manual work to supervision, quality control and exception handling. At the same time, companies need secure AI infrastructure: enterprise versions, access control, encryption and audit capabilities (for example, in a corporate environment), as well as DLP policies, data classification and usage logging. Without such foundations, organisations often halt AI usage altogether, which makes it harder to create genuinely new roles and upskill staff.

The state can reduce social costs through active retraining programmes and support for SMEs in accessing implementation consultancy and ready-made automation packages. Regulatory balance is also important: the EU AI Act and GDPR set the rules for AI use, especially in high-risk areas, so companies need clarity on “what is allowed and what is not” to prevent investments from stalling. Within organisations, tensions are reduced by sharing productivity gains (for example, bonuses, investment in development, creating new services) rather than treating AI solely as a tool for cuts. When redundancies are unavoidable, social dialogue and advance planning help: redeployment, training and outplacement programmes, before shock hits the team.

Future forecasts: scenarios for 3–10 years

The most realistic forecasts for 3–10 years include several scenarios in which AI sometimes speeds up work, sometimes strongly automates mass services, and in the most transformative variant partially autonomises task execution. In the “productivity calculator” scenario, changes remain moderate and employment shifts gradually, mainly through natural attrition and lower hiring. In a scenario of strong automation of mass services (BPO, customer service, content), bots and workflows handle large volumes of cases, while demand grows for oversight, analytics, quality control and relationships with key clients. In the AI agents scenario, part of the work can be carried out end-to-end, which strengthens the importance of control roles, security, policy design and accountability for outcomes.

Your own risk is best assessed by analysing the characteristics of the tasks you perform, rather than the job title itself. A set of five control questions helps here: can the output be easily evaluated by a metric, is the process repeatable, is the input data digital, does the company have a cost incentive to automate, and does someone need to bear legal responsibility. The more “yes” answers to the first four questions and “no” to the fifth, the greater the pressure to automate tasks. This diagnosis remains useful regardless of the industry, because it shows where it is worth shifting the emphasis towards quality, exception handling and decision-making.

Warning signs usually appear earlier than redundancies and concern the way the company is restructuring its operations. If an organisation is deploying RPA/LLM in a given area, reducing hiring, merging teams and introducing KPIs such as “time to handle per case”, it is often preparing the ground for a reduction in manual work. A second signal is often the centralisation of knowledge and the building of a bot for common questions, which gradually shifts the role of the people handling cases. By contrast, signals of opportunity include creating AI governance, a centre of excellence, data teams and automation roles, as well as tidying up processes and data, because then it is easier to own the area and build an advantage.

The impact of automation may be softened by demographic factors, because staff shortages in care, education or technical professions mean that AI is sometimes a tool for increasing the system’s “capacity”, rather than simply replacing people. At the same time, the risk of inequality and labour market polarisation is growing, because AI rewards people who can increase productivity, while placing ever greater pressure on routine roles. The best way to track trends is without the noise, by analysing requirements in job adverts, vendor announcements (Microsoft, Google, OpenAI), industry case studies and real implementations in a given sector and company. When the direction in job adverts and internal projects remains consistent, it is usually a stronger signal than media headlines.