AI is changing how tasks get done faster than it is changing how responsibility is shared.
Generative AI tools have now entered everyday work: writing drafts, summarising meetings, answering customer questions, and helping write code. The change is real, but uneven. Some tasks have truly changed, some have only changed tools, and some still demand human judgement.
What changes
The clearest change is in tasks that can be started from scratch quickly. Drafting a report outline, producing headline variations, summarising a long document, or turning scattered notes into a tidy structure can now be done in minutes. AI tools also help with initial research and exploring ideas.
The second change concerns the mix of skills. Workers are no longer asked only to produce text, but also to check, correct, and decide whether AI output is fit for use. The ability to give the right context and to recognise mistakes becomes more important than typing speed.
The third change is the expectation of speed. When part of the work can be sped up, deadlines and workloads can shift too. If this is not discussed, workers can get caught in rising expectations without support in the form of tools, training, or time to verify.
What stays
Responsibility stays with people. AI tools cannot bear the legal, reputational, or safety consequences of decisions that are made. AI output therefore needs to be treated as raw material, not a final decision.
Judging context also stays human. AI can produce answers that sound convincing without understanding the customer's situation, internal policy, or long-term consequences. Workers who understand the context are still needed to judge whether an answer is right.
Working relationships and trust do not change automatically either. Customers, partners, and colleagues still judge consistency, accountability, and clarity of communication. A new tool can speed up work, but it does not replace accountability.
What needs to be agreed
Before AI tools are widely used, organisations need to agree on some basics. First, which tools are allowed and for what types of data. Second, which data must not be entered into third-party tools. Third, who must check the output before it is used or sent to the public.
These agreements relate to personal data protection. Law Number 27 of 2022 on Personal Data Protection sets out the rights of data subjects and the obligations of data controllers. On the employment side, the applicable rules emphasise the employment relationship, obligations, and the protection of workers. Neither is technical guidance, but both indicate that the use of AI at work cannot be separated from the organisation's responsibility.
Internal policy should also cover training and reporting channels. Workers need to know where to ask when in doubt, and who has the authority to stop the use of a risky tool. Without clear channels, workers tend to use tools quietly or to avoid them altogether.
Who is affected
Workers in administration, customer service, media, and software development feel the change in tasks fastest. Managers need to adjust how they assess performance: not just the amount of output, but also the quality of verification and compliance with data policy.
Job seekers face a changing market. Postings may require the ability to use AI tools, but also the ability to explain their limits. More senior workers need to share their knowledge of context and risk with newer colleagues.
What is still unclear
The net effect on the number of jobs cannot yet be determined. Some tasks may disappear, some will change, and some will be new. Productivity measurement is also not yet uniform, so claims of "up by so many percent" need to be examined for their method.
It is also unclear how responsibility is divided when AI output causes an error. Is the user responsible, the team that built the tool, or the organisation that allowed it? The answer to this question is still developing and needs to be written into internal policy.
Closing
AI at work is safest to read as a change in tools and in the division of tasks, not as a wholesale replacement. What needs to be protected is clarity of responsibility, data protection, and the habit of verifying. If a claim about AI sounds too certain, ask: what data was used, who checked it, and who bears the consequences?