The Roles AI Is Really Changing — and the Ones Just Getting More Paperwork
Which jobs in Spain and the EU are actually being automated, which are quietly piling up oversight work, and what the WEF, Cedefop, IAPP and field data say about both.
In Spain, customer service employment kept growing through 2025 even as nearly every mid-size company in the sector deployed an AI agent on tier-1. The INE’s Encuesta de Población Activa shows the headcount has not fallen. The work has changed. Agents now spend more time auditing the bot’s answers than answering customers themselves.
That is the shape of AI in European labour right now. Not the wholesale elimination of jobs. A redistribution, with a quality-assurance bill attached. The data from 2025 and early 2026 is precise enough to draw a useful line between roles AI is automating and roles AI is burdening with new overhead.
For the task-level picture of what AI can technically automate, see What AI Actually Automates in 2026. For why most automation projects fail at the integration layer, see The No-Code Gap.
Key takeaways
- The WEF projects 92 million roles displaced by 2030 against 170 million created; clerical functions dominate the decline list (WEF Future of Jobs 2025, Jan 2025).
- Data entry keyers carry a 0.99 probability of computerisation, the highest of any knowledge-work occupation studied (Oxford Martin School, 2013).
- AI coding tools made code review harder, not easier: cloned code blocks rose 48% between 2020 and 2024 (GitClear, 2025).
- 98.5% of organisations expect to need additional AI governance staff within a year — oversight is the new growth function (IAPP, 2025).
Which roles face genuine automation risk
Clerical, repetitive, rules-based functions are bearing the brunt. The WEF’s 2025 Future of Jobs report projects administrative assistants and executive secretaries losing 6.1 million positions globally by 2030 — the second-largest absolute decline of any occupation category studied (WEF, Jan 2025). Accounting and payroll clerks: 1.65 million. Stock-keeping clerks: 2.64 million. Data entry clerks: 0.5 million.
In Spain and southern Europe the curve is steeper than the headline because the SME tissue is denser and back-office work is more concentrated. Cedefop’s 2025 European skills forecast has clerical occupations with the largest negative employment differential against the EU-27 baseline through 2030 (Cedefop, 2025).
Roles built around moving information from one system to another — filling forms, updating spreadsheets, logging transactions — are in structural decline. This is not prediction. For most of these functions the automation is already deployed and running.
The numbers go back further than recent headlines suggest. Frey and Osborne at Oxford Martin School assigned data entry keyers a 0.99 probability of computerisation in 2013, the highest of 702 occupations they evaluated (Oxford Martin School, 2013). What was a probability twelve years ago is closer to operational reality now.
Customer service: heavy automation, stable headcount
The most visible deployment in Europe right now is tier-1 customer service. Gartner expects agentic AI to autonomously resolve 80% of common customer service issues by 2029 (Gartner, Mar 2025). That figure is directionally consistent with what Spanish telcos, banks and utilities are putting into production.
What it does not predict is mass layoffs.
In a March 2025 industry survey, 95% of customer service leaders said they planned to keep their human agent workforce rather than reduce it. The reason is not sentiment. It is what the work has become. Erik Brynjolfsson, Danielle Li and Lindsey Raymond’s call-centre study at a large software firm found AI tools lifted overall productivity by 14%, with gains of 35% for the least experienced agents — two-month-tenure agents performing at the level of six-month-tenure ones (Brynjolfsson, Li & Raymond, NBER, 2023, n=5,179 agents).
That is not automation. It is augmentation, with a workload attached. Volume per agent went up. The complexity of what reached a human went up. Headcount stayed flat.
| Layer of customer service | Direction in 2025 |
|---|---|
| Tier-1 routine queries | Automated, no human in the loop |
| Tier-2 complex cases | Human + AI assist, productivity +14% to +35% |
| Tier-3 escalations and complaints | Human, with new audit-of-AI work attached |
| QA / supervisory roles | Growing — auditing whether the bot answered correctly |
The developer productivity paradox
Code generation tools — GitHub Copilot, Cursor, Gemini Code Assist — are among the most adopted AI tools in professional settings. The headline productivity numbers look good in isolation: faster completion, more features shipped, fewer context switches.
Harvard Business School’s consulting study found professionals using AI completed 12.2% more tasks, 25.1% faster, with output judged 40% higher quality (Dell’Acqua et al., HBS Working Paper 24-013, 2023). Real gains. The same paper also identifies a “jagged frontier” where AI fails unpredictably on tasks it appears capable of handling.
The downstream cost shows up in code review. GitClear analysed 211 million changed lines of code between 2020 and 2024 and found copy-pasted and cloned code blocks rose from 8.3% to 12.3% of all output, a 48% increase as AI coding tools spread (GitClear, 2025). Developers ship faster. Reviewers inherit more duplicated logic, more inconsistency, more surface area to check.
In Spain and southern Europe, where senior engineering profiles are scarcer than in the German-Dutch-Nordic axis, that compression has a second-order effect: the few seniors who can review AI-generated code become the bottleneck. Basic code generation gets easier. Review of that code gets harder. The role of Tech Lead expands by accident.
The roles gaining the most overhead from AI are usually the ones positioned directly downstream of the roles it is automating. — pattern observed across our 2025 audits in EU SMEs
How automating one role inflates another
What rarely surfaces in automation coverage is the function expanding because AI deployment requires human oversight.
The IAPP’s 2025 AI Governance Profession Report found that only 1.5% of organisations expected to need no additional AI governance staff within the coming year — near-universal demand for oversight roles. 23.5% cited difficulty finding qualified AI governance professionals as a top barrier (IAPP, 2025). In the EU, the AI Act’s high-risk obligations make this worse: companies in banking, insurance and healthcare have to staff these functions to deploy at all.
What tends to happen in practice — visible in three out of four projects we audited in 2025 — looks like this. A document processing function is automated. 80% of the volume moves to AI. What grows in its place:
- a review function to catch the 20% the AI misclassifies,
- an audit trail to satisfy DPO and AI Act conformity requirements,
- a governance process to manage model and prompt updates,
- a vendor-management workflow to track sub-processor changes.
Four new task categories in exchange for one routine. None of them existed on the org chart twelve months earlier.
The WEF projects 170 million new roles created by 2030 against 92 million displaced — a net gain of 78 million (WEF, Jan 2025). The net figure flatters the transition. New demand concentrates in technology and AI oversight. Displaced demand concentrates in clerical and administrative work. The two populations rarely share skills, geography or retraining path — and in Spain the geographical gap (Madrid and Barcelona vs. the rest) makes the mismatch worse.
What “changing” actually means for a role
The right question is not “is this role automatable?” It is “which tasks within this role are automatable, and what does the remainder look like?”
McKinsey’s November 2025 analysis is explicit: even roles with high technical automation potential still need people “to guide, supervise, and verify” (McKinsey Global Institute, Nov 2025). That phrase covers a specific kind of work: judging what good output looks like, catching failure modes the model does not flag, explaining AI decisions to people who have to act on them.
Where verification is straightforward and errors are cheap, oversight stays light. Where errors carry legal, financial or safety cost — regulated environments, customer-facing judgement calls, anything the AEPD or Banco de España can audit — oversight becomes a job in its own right.
A useful diagnostic: what does verification look like for this function? That question separates “changing” from “just getting more paperwork” better than any general automation percentage.
Map your own functions before the market maps them for you
The roles changing fastest are the ones where task-level automation is straightforward and the cost of an error is low. The roles accumulating new overhead are the ones where AI output has to be reviewed before anyone can act on it. Both patterns matter for planning. Neither shows up cleanly in headline projections.
The clearest next step is to map your functions against those two categories. Which tasks are routine enough to automate reliably? Which will generate verification work downstream? Which existing roles become oversight functions if the underlying task is automated?
Diagnose which tasks in your team can actually be automated — free at canihireanai.com.
Frequently asked questions
Which job functions face the highest documented AI automation risk in the EU?
Data entry, basic administrative support and tier-1 customer service carry the highest documented risk. Oxford Martin School assigned data entry keyers a 0.99 probability of computerisation. Cedefop’s 2025 forecast places clerical occupations with the steepest negative employment differential through 2030. WEF puts administrative assistants second in projected absolute job losses globally by 2030.
Is AI creating new jobs in Spain and Europe, or just eliminating old ones?
Both, but the populations do not overlap cleanly. WEF projects 170 million roles created against 92 million displaced by 2030. New demand concentrates in technology, green economy and AI oversight. Displaced demand concentrates in clerical work. In Spain the mismatch is sharper because new demand is geographically concentrated in Madrid and Barcelona while clerical decline is distributed across the rest of the country.
Why do developers still need to do code review if AI can generate code?
Because AI-generated code quality is declining in specific ways. GitClear found cloned and copy-pasted code blocks rose 48% as AI coding tools spread between 2020 and 2024. Generation volume went up. Coherence and uniqueness went down. Reviewers and architects who catch duplication and maintain system integrity are under more pressure, not less.
What does “new paperwork” from AI actually look like in practice?
When a function is partially automated, the roles downstream inherit oversight tasks: reviewing AI output before action, auditing decisions for compliance, managing prompts as models update, handling escalations the AI cannot resolve. IAPP data shows 98.5% of organisations expect to add AI governance staff within a year. The oversight function is growing. It is just not being hired for at the same speed.
The jobs gaining the most paperwork from AI are rarely the ones the headlines warn about. Not the data entry clerk. Not the tier-1 agent. The reviewer sitting downstream, now responsible for everything the model gets wrong. That is the shape of AI’s near-term impact on work in Spain and the EU. Not wholesale elimination. A redistribution, with a quality-assurance bill attached.