A workstation struggling during a video conference, an application server overwhelmed by automated requests, a security alert triggered by a misconfigured language model: this is the IT reality of 2024. This year’s tech trends are not just marketing announcements. They are changing how we size infrastructures, secure data flows, and manage regulatory compliance at the European level.
European AI Act: the regulatory timeline that changes IT projects
Before discussing tools or performance, we must address legal constraints. The European regulation on artificial intelligence (AI Act) came into effect on August 1, 2024. Its operational impact on companies’ IT systems remains underestimated in most annual tech assessments.
The implementation timeline is staggered. Since February 2, 2025, certain AI practices are strictly prohibited: large-scale social scoring, behavioral manipulation, certain forms of real-time biometric surveillance. Fines can reach 35 million euros or 7% of global turnover.
As of August 2, 2025, general-purpose AI models (notably large language models) are subject to technical documentation and transparency obligations regarding training data. By August 2, 2026, transparency obligations and a comprehensive sanctions regime will come into effect, along with a European AI office with oversight powers.
For IT teams, IT information on Code Web regularly details this type of regulatory topic applied to the digital sector. The question is no longer whether we use AI, but how we document and audit each software component that integrates it.
- Map all software components that include an AI model, including third-party libraries and cloud APIs
- Establish an internal compliance register with risk levels defined by the regulation (minimal, limited, high, unacceptable)
- Allocate a training budget for development and legal teams, as liability in case of AI errors remains a point of friction in court
Generative AI in production: moving beyond the gadget to ensure reliable usage
We’ve all seen spectacular demos of text, image, and code generation. In 2024, the real on-the-ground question is different: how to integrate generative AI into a production chain without creating technical debt or security vulnerabilities.
Concrete applications that hold up involve software development assistance, technical documentation generation, and automation of first-level support tasks. Feedback varies on this point, but teams achieving good results share a common trait: they limit the model’s scope of action to a specific business domain.
A language model connected to an internal knowledge base (RAG approach, Retrieval-Augmented Generation) produces more reliable responses than a generalist model queried without context. The difference is directly measurable in the factual error rate of the generated responses.
The shadow AI trap
Employees are using online generative AI tools without IT department validation. Customer data copied into a prompt, source code pasted into an external assistant: shadow AI poses a data leakage risk comparable to shadow IT from the 2010s. The solution lies in clear usage policies and approved internal alternatives.
Workstations and cloud: balancing local performance and remote resources
The sizing of desktop computers has changed. Cloud applications (SaaS, remote development environments, permanent HD video conferencing) consume more bandwidth and RAM than three years ago. A workstation configured in 2021 with 8 GB of RAM and an entry-level processor shows its limits on standard uses in 2024.
For office and browsing-oriented workstations, 16 GB of RAM and an NVMe SSD are now the functional minimum. For development or data processing workstations, we move to 32 GB without hesitation.
On the server and infrastructure side, the shift towards hybrid cloud is confirmed. Companies retain certain sensitive workloads on-site (critical business applications, regulated data) while outsourcing peak loads and testing environments. The decision hinges on three criteria:
- The acceptable latency for the applications used (a computer-aided design tool does not tolerate network round trips well)
- The actual cost of the cloud over three years compared to the depreciation of a physical server, including data egress fees
- The data localization requirements imposed by GDPR and now by the AI Act for high-risk systems
Cybersecurity in 2024: threats amplified by AI
Targeted phishing attacks have gained sophistication thanks to language models. An AI-generated phishing email no longer contains the spelling mistakes that used to serve as warning signals. Traditional anti-spam filters are no longer sufficient against contextualized messages that mimic the style of a colleague or supplier.
Audio deepfakes pose a real problem in phone validation procedures. Documented cases show attempts at CEO fraud using synthetic voices. The technical countermeasure involves systematic multi-factor authentication and double verification protocols for any transfer or sensitive access.
Passkeys and the gradual end of passwords
The adoption of passkeys (FIDO2 access keys) is progressing across major browsers and operating systems. This mechanism replaces the username/password pair with cryptographic authentication tied to the device. The migration to passkeys drastically reduces the risk of credential theft, but it requires compatible hardware and a deployment phase that takes several months in medium-sized organizations.
The IT world in 2024 is structured around two axes: complying with a European regulatory framework that leaves no room for improvisation, and adapting infrastructures to uses where AI is no longer an option but a software component among others, with its reliability and security requirements.
