Which IT tasks is AI actually changing?
By the RoleMath Editorial Team · Last updated 2026-06-19. Every figure traces to a cited source; we sell none of the options discussed. Draft pending human review.
Honest answer: AI is changing TASKS inside IT roles far more than it's eliminating whole roles. Researchers measure this as task 'exposure' - the overlap between what an occupation does and what AI can do - and it is NOT the same as job loss. In the most-cited study, Eloundou and colleagues (Science, 2024) estimate roughly 80% of U.S. workers have at least 10% of their tasks exposed, with about 19% at half or more. Below is what's most and least exposed across common IT work - framed as honest reasoning grounded in that research, not a per-role 'AI score' (no one can credibly publish one).
Key takeaways
- AI changes tasks, not whole roles - researchers measure task 'exposure' (overlap with AI), and exposure is NOT job loss.
- Routine, well-specified tasks (boilerplate code, first-draft queries, log triage, summarizing) are more exposed; judgment-heavy work (debugging unfamiliar systems, incident response, design, stakeholder translation) far less.
- Tools are used more to augment than to automate (~57% to 43% in Anthropic usage data) - and the most-exposed high-skill work is often the least automatable (OECD).
- There is no honest per-role 'AI exposure number' - we won't publish one; reason about your tasks and watch what certs and employers actually do.
What 'AI is changing tasks' actually means
The honest unit is the task, not the job. Researchers estimate 'exposure' - how much of an occupation's work overlaps with current AI capability. Eloundou and colleagues (Science, 2024) put roughly 80% of workers at 10% or more of tasks exposed, but they explicitly disclaim any forecast of how fast tools get adopted or whether jobs are lost. Two honest caveats travel with every exposure figure: exposure means a task can be sped up or assisted, not that the role disappears; and exposure and automation risk often run in opposite directions - the OECD (2023) finds high-skill, high-exposure occupations are frequently the least at risk of automation.
The kinds of IT tasks most exposed (reasoning, not a score)
Across IT work, the tasks most overlapping with current AI tools tend to be the routine and well-specified ones: generating boilerplate or repetitive code, drafting first-pass database queries or scripts, summarizing logs and documents, writing first-draft documentation, and triaging high-volume, low-ambiguity alerts. These are exactly where usage data shows AI assisting today - and notably, that usage skews toward augmentation (a person plus the tool) over full automation, roughly 57% to 43% in Anthropic's measure of how people actually use AI (usage data, not employment outcomes). This is reasoning about which tasks overlap with the tools, not a measured statistic about any one job - and it's why 'learn to work with AI tools' is more useful advice than 'avoid the exposed roles.'
What stays stubbornly human (so far)
The tasks least overlapping with current tools are judgment-heavy and context-dependent: debugging an unfamiliar system under pressure, leading an incident response, designing an architecture against real constraints and trade-offs, deciding what NOT to build, translating between stakeholders, and owning the accountability when something breaks. Security work is a clear example - AI helps both attackers and defenders, but the adversary adapts, so human judgment and response stay central. None of this is a guarantee about the future; it's a description of where today's tools are weakest, consistent with the OECD's finding that the most-exposed high-skill work is often the hardest to automate outright.
What this means for your plan
Don't try to dodge 'exposed' tasks - they exist in almost every role. Instead, build the judgment-heavy skills that current tools are weakest at, and get genuinely good at working WITH AI tools, since that's how the work is increasingly done. Watch the verifiable signals rather than predictions: what AI content the certifications you're considering now test (a fact on the vendor's page) and the occupation-level outlook (a cited forecast, not a guarantee). And ignore anyone offering a per-role 'AI exposure percentage' or an 'AI-proof' path - those numbers aren't sourceable, which is exactly why we don't publish them.
Frequently asked questions
Does AI replace whole IT jobs or just tasks?
Mostly tasks, not whole roles. Researchers measure task 'exposure' (overlap with AI), and exposure is not job loss - the authors of the most-cited study (Eloundou et al., Science 2024) explicitly disclaim any job-loss forecast. AI is changing how work is done inside roles more than eliminating roles wholesale.
Which IT tasks are most affected by AI?
Routine, well-specified ones: generating boilerplate code, drafting first-pass queries and scripts, summarizing logs and docs, and triaging high-volume low-ambiguity alerts. Usage data shows AI mostly assisting (augmentation) rather than fully automating these (~57% to 43%, Anthropic).
Which IT tasks does AI struggle with?
Judgment-heavy, context-dependent work: debugging unfamiliar systems, incident response, architecture and trade-off decisions, stakeholder communication, and accountability. The OECD (2023) finds the most-exposed high-skill work is often the least automatable - exposure and automation risk run in opposite directions.
Can you give me an AI exposure score for my role?
No - there's no honest, sourceable per-role 'AI exposure number,' so we won't publish one. What you can do is reason about which of your tasks are routine vs judgment-heavy, and watch verifiable signals: what AI content certs now test and the occupation outlook.
Related, with the cited detail
- Are IT certifications worth it?
- Will AI replace tech jobs?
- How to use AI to study for IT certifications
- Start the RoleMath planner
Sources
Figures in this article are cited to the sources named in the Citation Ledger below and on each linked cited page. This page stays draft_noindex pending human citation review.
Citation Ledger
| ID | Supports | Evidence | Source |
|---|---|---|---|
| CIT-01 | ~80% of workers have at least 10% of tasks exposed; exposure = task overlap, not job loss; authors disclaim adoption/job-loss forecast | Occupational LLM task-exposure estimates | Eloundou et al., 'GPTs are GPTs', Science 2024 |
| CIT-02 | Exposure and automation risk run in opposite directions; high-skill high-exposure work least automatable | AI exposure as task overlap; capability is not probability of automation | OECD Employment Outlook 2023 |
| CIT-03 | AI usage skews to augmentation over automation (~57% to 43%) | Measured usage of AI tools, not employment outcomes | Anthropic Economic Index |