The IT department is an interesting domain for AI, because it combines two kinds of work that respond in completely different ways to automation. On one side are processes that follow a fixed pattern: a new employee is given an account, a backup runs at night, a password is reset after a request. On the other side is work where the outcome depends on context, risk assessment and consequences that cannot be captured in a script. Our taxonomy counts 39 tasks within this domain, and the distribution across the three categories largely follows that dividing line.
This concerns tasks with a clear input, a fixed procedure and a verifiable outcome. Setting up the workplace, accounts and access rights for a new employee is a good example of this: the job profile determines which rights are needed, and that link can be automated without anyone having to think through the details each time. The same applies to revoking access upon termination of employment, where the risk of acting too late is greater than the risk of automation, and to performing and checking backups, where a system itself can signal whether a task has succeeded or not. These tasks fall into category 1 because the standard is fixed: there is a clear right and wrong, and the steps in between do not need to be assessed anew each time. This does not mean that no oversight is needed at all over the whole, but the execution of the task itself does not require human judgement case by case.
This category is large in the IT department, and that is no coincidence. Many tasks do have a recognisable pattern, but the outcome touches on risks that a person must weigh up. Periodically reviewing access rights is one such task: AI can flag discrepancies between role and current rights, but whether an exception is justified requires an assessor who knows the context and who approves or rejects it with reasons. Testing recovery from a backup falls into the same category: the testing itself can be automated, but the conclusion as to whether a system has actually been sufficiently restored requires a look that goes beyond a green checkmark. Setting up an integration between systems and monitoring and resolving a disruption in it also belong here: AI can speed up the build and detect anomalies, but the decision on how a disruption is resolved without damaging data remains a judgement someone must make. The pattern in this category is always the same: AI does the recognising and proposing, a person does the assessing and deciding.
A smaller part of IT work remains solidly human regardless of how mature AI becomes, because the core of the task does not consist of pattern recognition but of interpretation and accountability. Ad-hoc data analysis is an example of this: answering a specific question from the organisation requires that someone understands what is actually being asked, which data is relevant and which conclusion holds up. Checking and cleaning data quality touches on the same problem as soon as it goes beyond detecting duplicate records: the question of which source represents the truth in case of conflicting data requires knowledge of the organisation that is not contained in the data itself. Managing master data also remains partly human work where it concerns establishing definitions and agreements between departments, although the input itself is often automatable. These tasks remain human work not because AI cannot handle the technology, but because the responsibility for the outcome cannot be delegated to a system.
This classification concerns tasks, not roles and not individuals. An IT employee generally carries out tasks from all three categories, and which part of their task package changes depends on the precise makeup of their role. That a task falls into category 1 also says nothing about what an employer does with that: that is a choice with its own legal requirements, for example regarding employee participation and labour law, which falls outside the scope of this classification. What the classification does offer is a factual starting point: which tasks can already be supported with AI today, which require oversight, and which remain unchanged. Similar patterns can be recognised in other domains; for instance what work can AI take over in procurement shows that there too the dividing line often lies in the question of whether an outcome has a fixed standard or requires a judgement call, and what work can AI take over in logistics illustrates how a seemingly operational department still has many oversight tasks once errors become costly.
To see how many hours within an IT role can currently be assigned to one of the three categories, a concrete inventory is needed of the tasks someone actually performs, not just the job title. The free quickscan from ftetoai offers a first indication for this: twelve questions, no account required, resulting in an indication of what portion of the hours in the profile can be taken over by AI today. That is a starting point, not a final verdict. The full work scan, which goes deeper into individual tasks and the reasoning behind them, is still under construction. What a follow-up measurement shows after change, and how it compares to the initial inventory, can be read at what a remeasurement shows.
Vraag maar. Ik ken de kennisbank van deze site; wat ik niet weet, zeg ik erbij.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.