No time for AI becomes a staffing risk

5 min read

73 percent of surveyed employees use AI at work, yet only 24 percent have working time rules that include learning time. Since February 2025, Article 4 of the AI Act has required businesses of all sizes using AI to take measures supporting staff AI literacy. A lack of learning time therefore poses staffing and compliance risks.

Related:A year of mandatory electronic patient records: connected but barely used  /  Neuralink gives an ALS patient his voice back

Developers debate learning time

On 9 October, developer Geoffrey Huntley published a blog post with the brief title “replaced”, or “ersetzt” in German. For Huntley, using AI in software development is no longer optional. His advice to hiring managers covers three points: schedule development time, classify employees by AI literacy and distinguish between “producers” and “consumers”. Producers automate workflows with AI; consumers simply operate the tools. He provides no classification method.

A day later, software engineer Addy Osmani raised an objection. A team allowed to experiment with AI only after work receives a tacit message that using it does not count.

Advice to employees to quit if their employer bans AI shows how heated the debate has become. Michael Taiwo replied that AI literacy matters. But telling someone who pays rent for three people to quit is costly advice. The debate comes from the software industry. Its underlying question also applies to the Mittelstand: does learning time count as working time?

What is AI literacy? AI literacy means the knowledge and skills to use AI tools competently at work and assess their limitations. Article 4 of the AI Act requires providers and deployers to take measures supporting staff in developing this competence. Under the revised wording of 27 July 2026, they are expressly exempt from guaranteeing any individual’s proficiency level.

AI use and learning time compared

In the Bitkom Akademie 2026 training study, 73 percent of surveyed employees say they use AI at work, up from 64 percent the previous year. 79 percent want to improve their AI literacy. Yet only 24 percent have working time rules that accommodate learning time. Just 21 percent say their manager actively raises the subject of further training.

The survey covered 1,275 working people and 84 management team members in calendar weeks 17 to 22; it is not representative. Management respondents favour developing their own staff. 77 percent respond to changing skills requirements primarily by training existing employees, ahead of automation and recruitment. At the same time, 45 percent acknowledge that investment tends to favour technology over training.

A lack of time has long slowed digitalisation in the Mittelstand. In a representative survey by digital industry association Bitkom published in March, covering 604 companies with at least 20 employees, 41 percent use AI, up from 17 percent the previous year. 66 percent cite a lack of time as a barrier to digitalisation. Data protection requirements rank higher at 77 percent, followed by skills shortages at 70 percent. Learning suffers from the same time constraints.

What Article 4 requires

Article 4 of the AI Act has applied since 2 February 2025. Since 27 July 2026, it has applied as amended by the Omnibus Regulation (EU) 2026/1744. It requires providers and deployers to take measures supporting the development of their staff’s AI literacy. Deployers are companies using AI systems in their own operations. They are expressly exempt from guaranteeing any individual’s proficiency level.

According to guidance from the Bundesnetzagentur, Article 4 prescribes no specific training format. Documenting the measures is recommended. A lack of AI literacy may constitute a breach of the duty of care, particularly where harm occurs. This means the obligation to exercise ordinary care in business operations.

The obligation requires demonstrable measures without prescribing an expensive format. The main commitment is therefore the working time employees receive for learning.

Three steps for management

The first step is a fixed weekly learning slot, entered in the calendar like a client meeting. This removes the tacit message Osmani describes. Office staff might use the time to try AI for minutes or correspondence. Sales teams can work on draft proposals and meeting preparation, while specialists can explore analysis and research. The aim is to move from consumer to producer, automating one’s own workflows with AI.

The second step is an interview question. Huntley sees a dismissive attitude towards AI, undisclosed AI use during the interview and a lack of personal experience as warning signs. He sets a high baseline for developers. They should be able to explain at a whiteboard how a coding agent works: a program that writes code independently. This includes two concepts. The context window is the amount of text a model processes at once. Tokenisation means breaking text into small units. Personal projects are strong indicators, as are verifiable organisational changes for managers.

For office and sales roles, a simpler question about a personal example from everyday work is enough. Asking how the candidate checked the result is revealing. It addresses an imbalance that the account @bertlef_ warned about in the thread on Huntley’s post. Hiring only AI-focused people could harm a team because nobody would solve problems without a tool.

The third step is a brief record. It documents who participated in which measure and when, and which AI tools are in use. This follows the recommendation to keep a clear written record of the measures. So far, 51 percent of surveyed management teams have embedded further training in their strategy.

Frequently asked questions

Image source: AI-generated (October 2026)

Translated from the German original using artificial intelligence. The German version is authoritative.