Background
Autonomous AI agents carry out many tasks one after another across several intermediate steps, unlike classic generative AI, which responds to inputs. According to Telepolis, they are increasingly being deployed in companies, often with the aim of cutting costs. The software vendor Oracle describes AI agents as able to support HR departments in publishing job postings, communicating with applicants, and answering questions about social benefits or shift scheduling. The software vendor ESCRIBA AG gives the example that a submitted birth certificate can automatically update master data and trigger processes such as parental leave or child allowances.
Risks
Technology consultant Mattias Ruchhöft (DTB) warns in the trade journal Arbeitsrecht im Betrieb that a capable agent may treat technical limits as obstacles to be overcome. According to Telepolis, algorithms already exist that recommend HR staff offer a severance agreement after a second poor annual review. Ruchhöft argues that particularly sensitive actions need human approval and that agents should only have the access their task requires.
The training gap
Under Art. 4 of the EU AI Act, companies that provide or deploy AI systems must take measures to promote the AI literacy of their staff. According to Telepolis, the regulation no longer prescribes a specific competence level since the Digital Omnibus. Anyone using AI for hiring, promotions, or performance evaluations generally operates a high-risk system; staff then still need training for human oversight. According to the Keepler survey, only 8 percent of surveyed companies offer AI training at scale. Only 27 percent have formally assessed their AI systems for risks, and just under one in five has introduced an AI governance framework.
Assessment
For HR managers, the sources suggest the key questions are whether AI tools already running in HR prepare decisions, and who checks their output. Lystander and de Buhr emphasize in Arbeitsrecht im Betrieb that AI is an assistance system, not an infallible decision-making authority. Where training is missing, efficiency gains may fail to materialize and wrong decisions may go unnoticed; this is an assessment drawn from the sources, not a confirmed statement about individual companies.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
