Towards the end of the year we get more and more calls with the same question: “Does the AI Act actually affect us?” The short answer: very probably yes. The longer answer is more encouraging than many think.

What applies now — and what is coming

The EU AI Act has been in force since 2024 and is being phased in step by step: bans on certain practices and AI literacy obligations already apply, the rules for providers of large general-purpose models have applied since August 2025, and the big block of high-risk requirements follows in August 2026. For most companies this does not mean panic but taking inventory: which AI systems are in use, which risk class do they fall into, who is responsible, and how are decisions documented? Anyone who starts this stocktake in 2025 will enter 2026 relaxed.

Agents need rules — even without a law

Regardless of Brussels: the more autonomously systems act, the more important guardrails become. The principle of limited autonomy has proven itself in practice — agents may do a lot, but within clearly defined limits: graded permissions, approval thresholds for critical actions, complete logs and an emergency stop worthy of the name. That is not bureaucracy but simply professional operation. Nobody would think of granting a new employee full power of attorney on day one.

Trust becomes a selling point

The most exciting observation from our projects: clients who have set up their AI governance properly win tenders with it. “For every automated decision we can show you how it came about” is a sentence that opens doors in 2025 — with clients, works councils and regulators alike. Governance is not a cost block. It is the foundation on which automation can be scaled in the first place.

If you want to make your AI landscape AI-Act-proof without losing the joy of innovation: striking exactly this balance is our daily business.